Delft University of Technology

Solar home systems for improving electricity accessAn off-grid solar perspective towards achieving universal electrificationNarayan, Nishant

DOI10.4233/uuid:aa29b04f-4cd7-41fa-b48b-5edc75fef104Publication date2019Document VersionFinal published versionCitation (APA)Narayan, N. (2019). Solar home systems for improving electricity access: An off-grid solar perspectivetowards achieving universal electrification . https://doi.org/10.4233/uuid:aa29b04f-4cd7-41fa-b48b-5edc75fef104

Important noteTo cite this publication, please use the final published version (if applicable).Please check the document version above.

CopyrightOther than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consentof the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Takedown policyPlease contact us and provide details if you believe this document breaches copyrights.We will remove access to the work immediately and investigate your claim.

This work is downloaded from Delft University of Technology.For technical reasons the number of authors shown on this cover page is limited to a maximum of 10.

SOLAR HOME SYSTEMS FOR IMPROVINGELECTRICITY ACCESS

AN OFF-GRID SOLAR PERSPECTIVE TOWARDS ACHIEVINGUNIVERSAL ELECTRIFICATION

Dissertation

for the purpose of obtaining the degree of doctorat Delft University of Technology

by the authority of the Rector Magnificus, Prof. dr. ir. T. H. J. J. van der Hagen,chair of the Board for Doctorates

to be defended publicly on Wednesday, 20 November 2019 at 10:00 am

by

Nishant Shankar NARAYAN

Master of Science in Sustainable Energy Technology,Delft University of Technology, The Netherlands,

born in Dombivli, India.

This dissertation has been approved by the promotors.Composition of the doctoral committee:

Rector Magnificus chairpersonProf. dr. ir. P. Bauer Delft University of Technology, promotorProf. dr. M. Zeman Delft University of Technology, promotorDr. Z. Qin Delft University of Technology, copromotor

Independent members:

Prof. dr. P. D. Lund Aalto UniversityProf. dr. ir. A. H. M. Smets Delft University of TechnologyDr. ir. R. M. E. Valckenborg Eindhoven University of TechnologyDr. ir. J. C. Diehl Delft University of TechnologyProf. ir. M. A. M. M. van der Meijden Delft University of Technology, reserve member

This research was supported by a fellowship from TU Delft | Global Initiative, a program ofDelft University of Technology to boost science and technology for global development.

Keywords: energy access, SDG 7, solar home systems, solar energy, batteries, ruralelectrification, multi-tier framework, GIS, microgrids

Cover design: Sanaa Degani.

Printed by: Ipskamp Printing (https://proefschriften.net/)This thesis is printed on 100% FSC-certified paper.

ISBN 978-94-6366-217-8

Nishant Narayan, 2019

Except for the cover and where otherwise noted, this this work is licensed under a CreativeCommons Attribution–NonCommerical–ShareAlike 4.0 International License. To view acopy of this license, visit http://creativecommons.org/licenses/by-nc-sa/4.0/.

An e-version of this dissertation is available at http://repository.tudelft.nl/

To Appa, who always gave back to others. Wish you were here.

To Mathe. I wish I could be as selfless as you. I try.

To Sanaa, my love. I know you are happier than me as I finish this book. You’ve enduredthrough this more than I could ever have asked for. On the bright side, now there’s one

less PhD project standing between me and the chores.

CONTENTS

Summary xi

Samenvatting xiii

1 Introduction 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Energy access 11.1.2 Illuminated but not electrified 41.1.3 Challenges with grid-based electrification 51.1.4 Solar Home Systems 7

1.2 Limitations of SHS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71.2.1 Cost 81.2.2 Battery in an SHS 81.2.3 Optimal system sizing 81.2.4 Load demand in off-grid systems 91.2.5 Power availability 91.2.6 Climbing up the electrification ladder 9

1.3 Scope, Objective and Research questions . . . . . . . . . . . . . . . . . . 111.3.1 Scope of this dissertation 111.3.2 Main objective 111.3.3 Research Questions 11

1.4 Research publications. . . . . . . . . . . . . . . . . . . . . . . . . . . . 111.5 Dissertation layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 The long road to universal electrification: A critical look at present pathwaysand challenges 192.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.1.1 Multi-tier framework for measuring electricity access 202.2 Pathway 1: Grid extension . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.2.1 Competition with off-grid renewables 232.3 Pathway 2: (Off-grid) centralized microgrids and mini-grids . . . . . . . . 25

2.3.1 Clarification in terminology 252.3.2 Centralized microgrids 252.3.3 Microgrids VS. grid extension 262.3.4 Microgrids VS. standalone SHS 272.3.5 Disadvantages of centralized microgrids 29

2.4 Pathway 3: Solar-based standalone systems. . . . . . . . . . . . . . . . . 302.4.1 SHS VS. grid and microgrids 30

v

vi CONTENTS

2.5 Present issues with SHS-based electrification . . . . . . . . . . . . . . . . 322.5.1 Climbing the electrification ladder 322.5.2 The paradox of SHS-based electrification - Watt’s the matter! 33

2.6 The holy grail of universal electrification . . . . . . . . . . . . . . . . . . 352.6.1 SHS-based microgrid: A means to get there? 362.6.2 Comparison of the various electrification pathways 38

2.7 Challenges for the SHS-based electrification vision . . . . . . . . . . . . . 392.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 Load profile construction 433.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.1.1 Multi-tier framework for household electricity access 443.1.2 Off-grid appliances 443.1.3 Importance of Load profiles 453.1.4 Need for load profile construction 463.1.5 Highlights 46

3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.2.1 Literature review 473.2.2 Load profile parameters 493.2.3 Types of appliances 50

3.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503.3.1 Load classification 513.3.2 Model parameters 523.3.3 Stochastic load profile model 573.3.4 Advantages of the methodology 58

3.4 Results and Discussions. . . . . . . . . . . . . . . . . . . . . . . . . . . 593.4.1 Stochastic load profiles for MTF 593.4.2 Load profiles: Main parameters 613.4.3 Implications on system design 633.4.4 Comparison with field data 63

3.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4 Estimating battery lifetime in Solar Home Systems 694.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.1.1 Literature study 714.1.2 Contributions of this chapter 73

4.2 Battery lifetime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.2.1 Battery parameters 734.2.2 Causes of battery degradation 74

4.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754.3.1 Battery data from the manufacturer 754.3.2 SHS application and load profile 774.3.3 Overall battery usage model 784.3.4 Dynamic capacity fading model 81

CONTENTS vii

4.4 Results and discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . 834.4.1 Battery Usage 834.4.2 Lifetime estimation 854.4.3 Comparison with an empirical battery lifetime estimation model 864.4.4 Relevance for SHS design 88

4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

5 Exploring SHS boundaries for electrification: Optimal SHS sizing 935.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

5.1.1 Importance of optimal SHS sizing 945.1.2 Literature study 955.1.3 Contributions of this chapter 96

5.2 System metrics and parameters . . . . . . . . . . . . . . . . . . . . . . . 965.2.1 System metrics 965.2.2 System parameters 97

5.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 985.3.1 Inputs to the SHS model 985.3.2 Modular SHS architecture 995.3.3 Dynamic PV output 1005.3.4 Estimating battery lifetime 1015.3.5 Power management scheme for standalone SHS 1015.3.6 Converter rating 1025.3.7 Multi-objective optimization for standalone SHS sizing 103

5.4 Results and discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . 1065.4.1 Dependence of SHS parameters on size 1065.4.2 Multi-objective optimization for SHS sizing 108

5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

6 Quantifying the benefits of SHS-based DC microgrids 1176.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

6.1.1 Literature study 1186.1.2 Contributions of this chapter 120

6.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1206.2.1 Location and meteorological inputs 1206.2.2 Stochastic load profiles 1216.2.3 System metrics and parameters 1216.2.4 Optimal standalone SHS sizes for the MTF 1226.2.5 SHS interconnection-based DC microgrid 122

6.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 1286.3.1 Energy exchange in the SHS-based microgrid 1296.3.2 Comparison of battery charging using excess energy 1296.3.3 Impact of microgrid size 1306.3.4 Benefits of microgrid on SHS sizing 133

viii CONTENTS

6.4 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

7 Optimal microgrid layout using Geographic Information System (GIS) 1397.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

7.1.1 Literature review 1407.1.2 Scope of this study 1437.1.3 Contributions of this chapter 143

7.2 Graph theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1447.2.1 Graph theory: some concepts and definitions 1447.2.2 Trees in graph theory 1467.2.3 Graph theory applied to network analysis 148

7.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1497.3.1 GIS-based data processing 1497.3.2 Microgrid topology creation 1527.3.3 Layout optimization 154

7.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 1607.4.1 Topology creation using QGIS 1607.4.2 Layout optimization 162

7.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1667.5.1 Recommendations and Future work 166

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

8 Conclusion 1698.1 General conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1698.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1708.3 Topical conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

8.3.1 Electrification pathways 1708.3.2 SHS 1718.3.3 SHS-based DC microgrid 172

8.4 Recommendations and future work. . . . . . . . . . . . . . . . . . . . . 1728.4.1 SHS 1728.4.2 SHS-based microgrids 1738.4.3 Multidisciplinary research 174

Appendix A: Constructing Equivalent Electrical Circuit Models of Battery Cells 175A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

A.1.1 Selecting the Battery Model 175A.1.2 Importance of Low Current Battery Models for SHS 176A.1.3 Contributions 176

A.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176A.2.1 Battery Parameters 176A.2.2 Construction of the Dynamic Battery Model 177A.2.3 Storage Circuit 178A.2.4 Electrical Response Circuit 178A.2.5 Parasitic Reaction Circuit 180

CONTENTS ix

A.3 Methods and Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . 180A.3.1 Choice of Operational Variables 180A.3.2 Overall Methodology 181A.3.3 Equipment and Materials 181A.3.4 Storage Circuit 182A.3.5 Voltage Response Circuit 182A.3.6 Parasitic Branch 182A.3.7 EECM Construction 183

A.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 184A.4.1 Experimental Results 184A.4.2 Simulation and Validation 191

A.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

Appendix B: Evaluating the impact of temperature on SHS 199B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199B.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

B.2.1 Physical effects of temperature on SHS components 199B.2.2 Scope of this study 200

B.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201B.3.1 Inputs to the SHS simulation 201B.3.2 Estimating PV yield 202B.3.3 Battery sizing 205B.3.4 Assessing temperature impact on battery lifetime 205

B.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 206B.4.1 PV results 206B.4.2 Battery results 208

B.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

Appendix C: Decentralized Control-Scheme for DC-Interconnected SHS 211C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

C.1.1 Chapter layout 212C.2 Microgrid case-study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

C.2.1 SHS-based microgrid 212C.2.2 Scenario for microgrid simulation 213C.2.3 Input data to the microgrid model 213

C.3 Control of interconnected SHS batteries . . . . . . . . . . . . . . . . . . 213C.3.1 The need for dedicated battery control 213C.3.2 Battery Converter 214C.3.3 PV Converter 216C.3.4 Load Converter 216C.3.5 Principle of control for the 48–350 V converter 217

C.4 Simulation results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217C.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

x CONTENTS

List of Publications 223

Acknowledgements 227

About the Author 233

SUMMARY

To achieve great things, two things are needed:a plan, and not quite enough time

Leonard Bernstein

More than a billion people globally lacked access to electricity in 2016. For variousreasons, national grid extension is not an economically viable solution for the un(der-)electrified regions. As most of these electricity-starved regions lie in tropical latitudeswith abundant sunshine, the use of off-grid, solar-based solutions like solar home systems(SHS) is a logical approach, especially when considering the falling costs of photovoltaic(PV) module and battery storage in recent times. However, state-of-the-art SHS is limitedin its power levels and availability. Moreover, sub-optimal system sizing of SHS leads toeither over-utilization — and therefore, faster battery degradation — or under-utilizationof the SHS battery, leading to higher system costs. Additionally, off-grid SHS designssuffer from a lack of reliable load profile data needed as the first step for an off-grid PVsystem (e.g., SHS) design. The work undertaken in this dissertation aims to analyze thetechnological limits and opportunities of using SHS in terms of power level, availability,and battery size, lifetime for striving towards universal electrification.

ELECTRIFICATION PATHWAYS

In Chapter 2, the three main electrification pathways, viz., grid extension, centralizedmicrogrids, and standalone solar-based solutions like pico-solar and SHS are analyzed fortheir relative merits and demerits. Grid extension can provide broad-scale electricity andhigh power levels; however, it needs a certain level of population density and electricitydemand to be an economically viable pathway. Centralized microgrids also require aminimum electricity demand threshold and good knowledge of the expected electricitydemand before they are setup. Standalone systems like solar lanterns and SHS have, de-spite having the highest per Wp system costs, the highest adoption rates at the householdlevel, despite their limitation in power levels. The requirements from an ideal pathwayare also discussed, which must not only allow the use of higher power appliances, butalso enable the inevitable climb up the energy ladder for the consumer.

LOAD PROFILE CONSTRUCTION

Load profiles are often the first step for the technical design of an off-grid energy systemlike SHS, while reliable load profiles are often unavailable. In Chapter 3, a methodologyis presented to quantify the electricity demand of the households in the form of loadprofiles for the various tiers of the multi-tier framework (MTF) for measuring householdelectricity access. The so-called super-efficient off-grid DC appliances are considered inthis study. The described methodology is scalable, and the resulting load profiles can bemade more accurate with complementary field surveys for the target communities.

xi

xii SUMMARY

BATTERY AND SHSBattery is a critical component of the SHS; it is the most expensive SHS component whilealso the component with lowest lifetime. In Chapter 4, a non-empirical battery lifetimeestimation methodology is presented that can be used at the design phase of SHS forcomparing the performance of candidate battery choices at hand in terms of batterylifetime. An SHS case-study simulation for a tier-3 load is carried out and the batteryactivity is analyzed. Comparison of this proposed dynamic model with an experimentallyderived empirical model of LiFePO4 battery yielded very close results, with the state ofhealth (SOH) values over time being within less than 3% of each other.

An optimal standalone system size is calculated for each tier of energy consumptionin Chapter 5, taking into account the battery lifetime, temperature impact on SHS perfor-mance, power supply availability in terms of the loss of load probability (LLP), and excessPV energy. A genetic algorithm-based multi-objective optimization is performed, givinginsights on the delicate interdependencies of the various system metrics like LLP, excessPV energy, and battery lifetime on the SHS sizing. This exercise concludes that meetingthe electricity requirements of tiers 4 and 5 level of electrification is untenable throughSHS alone.

SHS-BASED MICROGRIDS

Consequently, a bottom-up DC microgrid born out of the interconnection of SHS isexplored in Chapter 6. A modular and scalable architecture for such a bottom-up, in-terconnected SHS-based architecture is introduced, and the benefits of the microgridover standalone SHS are quantified in terms of lower battery sizes and the defined systemmetrics. On modeling the energy sharing between the SHS, it is shown that battery sizinggains of more than 40% and 30% could be achieved with SHS interconnectivity at tier 5and tier 4 levels, respectively, as compared to standalone SHS to meet the same poweravailability threshold.

Finally, a Geo-Information System (GIS)-based methodology is presented in Chapter 7that takes into account the spatial spread of the households while utilizing graph theory-based approaches to arrive at the optimal microgrid topology in terms of network length.A total of 42 different remote sites around the world are considered in the study. Graphtheory-based layouts (minimum spanning trees) are seen to outperform conventionaltopologies like ring, spider, bus in terms of average network length, highlighting theusefulness of this study.

The research carried out in this dissertation underlines the technological limitations ofSHS in aiming towards universal electrification, while highlighting the benefits of movingtowards a bottom-up approach in building DC microgrids through interconnected SHS,which can enable the climb up the so-called electrification ladder.

SAMENVATTING

In 2016 hadden ruim miljard mensen wereldwijd geen toegang tot elektriciteit. Om meer-dere redenen is uitbreiding van het landelijke elektriciteitsnet voor deze contexten meestalgeen economisch verantwoorde oplossing. De elektriciteit-arme regio’s liggen voorna-melijk in de tropische gebieden, met veel zon, hierdoor zijn de off-grid, zonne-energieoplossingen, zoals solar home systems (SHS) een logische keuze, des te meer gezien deexponentieel dalende kosten van PV en batterijopslag in de laatste jaren. Echter heeft destate-of-the-art SHS belangrijke beperkingen, zoals relatief lage vermogens die ze kun-nen leveren en de beperkte beschikbaarheid van elektriciteit. Verder heeft suboptimaledimensionering van SHS negatieve gevolgen, in de vorm van of over-utilisatie oftewelsnelle batterij degradatie, of onder-utilisatie oftewel een te dure systeem. Ten slotte is ergebrek aan betrouwbare energieverbruiksprofiel data een belangrijke beperking in eenoptimaal ontwerp van een SHS. Dit proefschrift streeft ernaar om technische limieten enmogelijkheden van SHS, vooral vermogensniveau, beschikbaarheid, batterijcapaciteitdimensionering en levensduur te onderzoeken.

ELEKTRIFICATIE ROUTES

In Hoofdstuk 2 zijn de drie belangrijkste elektrificatie routes geëvalueerd: het landelijkeelektriciteitsnet, gecentraliseerde microgrids en standalone zonne-energie oplossingenzoals pico-solar en SHS. Uitbreiding van het elektriciteitsnet biedt toegang tot elektriciteitop schaal en hoge vermogenniveaus. Aan de andere kant zijn er bepaalde bevolkings-dichtheid en elektriciteitsverbruik nodig om deze oplossing economisch rendabel temaken. Voor een economisch verantwoord ontwerp van gecentraliseerde microgridsis eveneens een bepaald elektriciteitsverbruik vereist. Standalone oplossingen, zoalszonne-energie lantaarns en SHS, hebben een brede grote marktadoptie laten zien, al zijnze relatief duurder en beperkt in vermogen die ze kunnen leveren. De eisen van een idealeelektrificatie route zijn ook uitgestippeld, die zowel het klimmen van de zogenaamdeelektrificatieladder als gebruik van hoog-vermogen huishoudelijke apparaten mogelijkmaakt.

CREËREN VAN HET ENERGIEVERBRUIKSPROFIEL VAN HUISHOUDENS

Om tot een passend ontwerp van een off-grid zonne-energie systeem te komen, is hetvan groot belang om het verbruikspatroon van een huishouden te kennen, welke appa-raten van elektriciteit voorzien moeten zijn en op welke tijden gedurende de dag ze ingebruik zijn. Echter zijn betrouwbare energieverbruiksprofielen in doelregio’s vaak nietbeschikbaar. In Hoofdstuk 3 is een methodologie ontwikkeld om de elektriciteitsvraagvan huishoudens te kwantificeren in de vorm van energieverbruiksprofiel behorend totverschillende rangen van de multi-tier framework (MTF). MTF is geïntroduceerd omverschillende niveaus van huishoudelijk elektriciteitstoegang te duiden. Dit onderzoek

xiii

xiv SAMENVATTING

gebruikt zogenaamde super-efficiente off-grid DC huishoudelijke apparaten. De ontwik-kelde methodologie is schaalbaar en de nauwkeurigheid van de energieverbruiksprofielenkan met aanvullende veldonderzoeken verbeterd worden.

BATTERIJ EN SHSBatterij is een kritische schakel in een SHS, het is de duurste en de kortste levensduurcomponent in het systeem. In Hoofdstuk 4 is een methodologie geïntroduceerd om op eennon-empirische manier de levensduur van de batterij te berekenen. De methodologie kanin het SHS ontwerp gebruikt worden om de levensduur van verschillende batterijkeuzesmet elkaar te vergelijken.

Een simulatie voor een representatieve SHS voor MTF Tier 3 is uitgevoerd om debatterijactiviteit te analyseren. De dynamische model en de experimentele empirischemodel van de LiFePO4 batterij leveren goede resultaten overeenkomst, het verschil tussende state-of-health (SOH) van de batterij tussen de twee modellen is minder dan 3

Hoofdstuk 5 presenteert optimale dimensionering van een SHS voor elke MTF Tier.Daarbij is rekening gehouden met de levensduur van de batterij, de invloed van detemperatuur op de SHS, elektriciteit beschikbaarheid, loss of load probability (LLP), enzonne-energie overschot. Met behulp van op genetisch algoritmes gebaseerde multi-objective optimalisatie is inzicht gekregen in hoe de verschillende afwegingen tussen deLLP, zonne-energieoverschot en batterijlevensduur effect hebben op dimensionering vanSHS. Ten slotte is aangetoond dat de hoogste niveaus van MTF, Tiers 4 en 5, met SHSalleen niet haalbaar zijn.

SHS-GEBASEERDE MICROGRIDS

Aansluitend op de inzichten van Hoofdstuk 5, is een bottom-up DC microgrid bestaand uitmeerdere, onderling verbonden SHS in Hoofdstuk 6 verder onderzocht. De voordelen vandeze modulaire, schaalbare, SHS-gebaseerde microgrids vergeleken met standalone SHSwat betreft de batterijcapaciteiten en andere systeemindicatoren zijn gekwantificeerd.Het is aangetoond dat doordat elektriciteit tussen huishoudens gedeeld kan worden, voordezelfde elektriciteitsbeschikbaarheid voor MTF Tier 4 en Tier 5, respectievelijk 30% en40% minder batterijcapaciteit gebruikt kunnen worden. Tenslotte wordt in Hoofdstuk7 een op Geo-informatie systeem (GIS) gebaseerde methodologie geïntroduceerd dierekening houdt met de ruimtelijke verspreiding van de huishoudens en de grafentheoriegebruikt om tot — vanuit het oogpunt van netwerklengte — optimale microgrid topologiete komen. In totaal zijn 42 verschillende afgelegen locaties in de wereld gebruikt als case-studies. De lay-outs die uit de grafentheorie (minimale opspannende bomen) voortkomenblijken vanuit het oogpunt van de netwerklengte betere resultaten op te leveren dan deconventionele topologieën zoals ring, spider, bus.

Het onderzoek uitgevoerd in dit proefschrift brengt de beperkingen van het SHS-concept in het bereiken van universele elektriciteitstoegang naar voren. Tegelijkertijdwordt de nadruk gelegd op de voordelen van een bottom-up benadering van het bouwenvan DC-microgrids door het verbinden van individuele SHS. Daardoor kan de elektrifica-tieladder sneller en beter beklommen worden.

1INTRODUCTION

We are the first generation that can put an end to povertyand we are the last generation that can put an end to climate change

Ban Ki-moon

1.1. MOTIVATION

1.1.1. ENERGY ACCESSIn 2016, the 17 Sustainable Development Goals (SDGs) of the 2030 Agenda for SustainableDevelopment came into force. In the same year, an estimated 1.1 billion people globallylacked access to electricity [1].

Figure 1.1: Map showing the global population in millions without electricity access in 2016. (Obtained fromthe electricity access database of the IEA [2].)

1

1

2 1. INTRODUCTION

It is no wonder then that the United Nations defines SDG 7 as “Ensure access toaffordable, reliable, sustainable and modern energy for all” [3]. Specifically, SDG 7-1targets universal access to reliable, affordable, and modern energy services by 2030,for which increasing the global access to electricity is crucial [3]. Figure 1.1 shows theglobal distribution of the population lacking electricity access as of 2016. Majority of thispopulation lives in sub-Saharan Africa and South/South-East Asia.

Figure 1.2 depicts the 17 SDGs of the 2030 Agenda for Sustainable Developmentadopted at the United Nations. SDG 7 deals with energy, which will be the focus of thework described in this dissertation.

Figure 1.2: The 17 Sustainable Development Goals adopted in September 2015, set to be achieved by 2030(Image source: [3]). Focus of this dissertation will be on SDG 7.

BENEFITS OF ELECTRIFICATION

There is ample evidence reported in the literature on the many benefits electrificationbrings about across geographies. For example, health benefits can be reaped from electri-fication when incumbent household fuel such as wood, coal, and kerosene are replaced,as evidenced from a study in South Africa [4]. Clean energy sources like solar photovoltaic(PV), which is a technology that converts incident radiation into electricity, can providea viable alternative instead of dirty fuels like coal and kerosene. Switching over fromkerosene lamp to solar PV-based lighting has shown to have improved the health andincreased the human development index (HDI) in studies conducted in India [5]. Ingeneral, studies have shown a high degree of correlation between HDI and access toelectricity [6, 7]. Other health benefits come in the form of better nutrition and well-being;when moving from dirty fuels to solar-based lighting, savings from light expenditures arereportedly spent on better-balanced diet [8]. Even the advent of basic lighting servicesimproves not only the productivity at home, but also creates more opportunities for

1.1. MOTIVATION

1

3

women and improves the study hours of children [8]. The most direct and immediatebenefits are in the form of the technological leap and associated cost savings while goingfrom fossil-fuel-based indoor energy consumption to electricity. For instance, for a givenamount of light output, kerosene costs 325 times the cost of electrically lighting an incan-descent bulb, 1625 times that of a compact fluorescent lamp, let alone the more efficientoption like the LED lights [9, 10].

Thus, it can be seen how SDG 7 is inseparably interlinked to other SDGs, like SDG3 (good health), 4 (quality education), 5 (gender equality), 13 (climate action), amongothers (Figure 1.2).

THE ILL-EFFECTS OF ENERGY POVERTY

Contrasting the benefits of electrification are the ill-effects of lack of electricity andenergy, also sometimes referred to as energy poverty. Energy poverty is inseparablylinked to economic poverty. The most affected victims of energy poverty are usually theeconomically marginalized population of any country 1. Energy poverty necessitates thereliance on dirty fuels and often a high percentage of the household income expenditureon inefficient energy sources [9, 11]. Energy poverty can lead to serious health concerns(indoor air pollution, lack of efficient medical care), and also further fuel economicpoverty while impacting gender roles and educational opportunities. Additionally, thereare other environmental effects like greenhouse gas (GHG) emissions, which are usuallynot associated with smaller, economically poorer countries because they use relativelylittle commercial energy. However, almost 80% of total energy use in Africa is in the formof burning wood and other biomass fuels [12].

14.21.5

1.4

3.9

5.6 0.2 26.9

K KL BT C MPC PS Total

Figure 1.3: Estimated annual spending in $ billion on off-grid lighting and phone charging in Asia and Africain 2014. K=Kerosene, KL=Kerosene Lamps, BT=Battery Torches, C=Candles, MPC=Mobile Phone Charging,PS=Pico solar (small solar products like solar lanterns). Data sourced and adapted from [13].

1also known as Base-of-Pyramid (BoP), which has been defined in the past as a demographic representation ofthe inequality of income or wealth

1

4 1. INTRODUCTION

The irony of energy poverty is that the economically poor tend to pay more for thesame energy service as compared to the same energy services in the richer parts of theworld. A case in point is mobile phone charging. The estimated annual spending onphone charging and off-grid lighting per source in 2014 is shown in Figure 1.3. In 2014,the 240 million strong mobile phone subscriber base (living off-grid in Asia and Africa)were estimated to have spent almost $0.15 to $0.25 per phone charge 2, leading to anincredible per kWh equivalent cost of $30 to $50 [13]. Therefore, energy poverty can evenbe seen as one of the factors in keeping the energy-poor locked into the cycle of povertyand marginalization.

ENERGY ACCESS AND ELECTRICITY ACCESS

It should be noted that electricity access is a part of the broader category of energy ac-cess that additionally includes, e.g., energy for heating and cooking. As electrificationis already a crippling global problem that needs urgent attention, the work described inthis dissertation pertains to electrification only. Furthermore, cooking has an additionalfactor of strong cultural dimension, which prevents the applicability of a universal solu-tion. On the other hand, for electricity access, the benefit of scale could be potentiallyreached. Therefore, the term electricity access and energy access may have been usedinterchangeably in this dissertation while referring to household-level electrification.

1.1.2. ILLUMINATED BUT NOT ELECTRIFIEDFigure 1.4 shows the global horizontal irradiance for different regions around the world.A quick comparison with Figure 1.1 reveals the paradoxical nature of the regions inSouth/South-East Asia and sub-Saharan Africa: these are places amply illuminated by thesun, but un(der)-electrified. That is, despite lacking sufficient electricity infrastructure,these regions are blessed with abundant sunshine, making solar PV power the perfectcandidate source of electricity.

Figure 1.4: Map showing the global horizontal irradiance around the world in terms of daily and yearly sums.(Obtained from the Global Solar Atlas, owned by the World Bank Group and provided by Solargis.)

2this can amount to up to 10% of the daily household income for a BoP household!

1.1. MOTIVATION

1

5

FALLING SOLAR PV COSTS

The technology costs of the so-called exponential technologies like solar PV, LED andbatteries have been declining. The global average levelized cost of electricity (LCOE) forutility-scale PV in 2017 reduced to $0.1/kWh, representing a 73% fall between 2010 and2017 [14]. Figure 1.5 shows the LCOE range for utility-scale PV from 2010-2017. Solar PV isincreasingly competing neck-and-neck with conventional fossil fuel-based power sources,and that too without any subsidies; by 2020, the average LCOE for solar PV is poised tofall below $0.06/kWh [15]. These trends not only make solar PV the candidate renewableenergy technology of choice but also make it the opportune moment to accelerate solarPV-based electrification efforts, in line with the idea of “clean energy” of SDG 7.

Figure 1.5: Global weighted average and range of LCOE from utility-scale solar PV projects from 2010-2016(Source: [15]. Data from IRENA Renewable Cost Database [15]).

1.1.3. CHALLENGES WITH GRID-BASED ELECTRIFICATIONFigure 1.6 shows the distribution of the global population without electricity acrossregions. A vast majority (nearly 85%) of the global population without electricity accesslives in rural areas.

Due to the remote location of the unelectrified villages and unstable electricity-grid inmany of these regions, grid-based electrification is certainly not an immediate solution toeradicate energy poverty. Despite accelerated efforts towards increasing energy access,

1

6 1. INTRODUCTION

the current rate is still deemed insufficient to meet SDG 7 by 2030 [16]. Moreover, thereare many factors why grid-based electrification has not reached uniformly across thewhole country for several nations. Examples of these factors include inefficient policies,corruption, technological inefficiency, among others [17, 18, 19].

Grid-based electrification for remote rural areas is in general fraught with financialrisks. For example, 55% of Kenya Power and Lighting Corporation (KLPC)’s customers,who are situated mainly in rural areas, spend roughly $3 a month on electricity [20]. Evenfor higher power consuming customers, it has been shown that the payback period ona typical KPLC rural grid connection is over 44 years [21]. Thus, in many areas, as morerural customers are added to African utilities, the utilities tend to lose more money.

The other problem plaguing the grid-based electricity in the un(der-)electrified re-gions of the world is the reliability of power supply. It was observed in 2018 that the gridconnection in urban and peri-urban areas in the Africa cities of Dar es Salaam and Nairobisupplied reliable power within the standard usable voltage range for only 47 percent ofthe time [21].

Transmission and connection costs are also bottlenecks for grid-based electricity.Up to 40% of all grid-based electricity costs can be attributed to transmission costs[10]. Economically viable grid extension demands a certain population density and perconnection electricity consumption, which are both factors that are almost certain to beworse for the least grid-connected areas of any country. Moreover, average grid extensioncosts can be high, estimated to be at around $500 per connection [10].

Africa Asia ME+LA0

200

400

600600

439

34

Regions

Pop

ula

tio

n[m

illi

on

s] Total Rural

Figure 1.6: Distribution of globally unelectrified population as of 2016 (data sourced from IEA [22]). ME:Middle-East; LA: Latin America.

Finally, a grid-based electricity paradigm is still fossil-fuel-based in many countries,leading to worsening effects on climate change while exhibiting technology-inertia interms of switching to or integrating a higher proportion of renewables.

In the presence of these challenges facing grid-based electrification, the other alter-native is off-grid electrification. The biggest advantage of off-grid electrification is thataccess to basic energy needs can be accelerated in comparison with grid-based electrifi-cation. Additionally, it is far easier to integrate renewables in off-grid systems due to the

1.2. LIMITATIONS OF SHS

1

7

opportunity to design the systems from scratch as well as their relatively smaller sizes.Moreover, for last mile connectivity and reaching low population density areas with lowelectricity demand, off-grid solutions are the most viable options.

Note: A detailed review of the challenges of grid-based electrification in comparisonwith off-grid electrification can be found in Chapter 2.

1.1.4. SOLAR HOME SYSTEMS

Solar PV-based off-grid rural electrification seems to complement, as well as compen-sate for, the grid extension efforts in most of the target regions in need of acceleratedelectrification. Solar Home Systems (SHS) are the perfect examples of solar PV productsproviding off-grid electrification. An SHS is usually defined as a solar PV-based generatorrated 11 Wp to more than 100 Wp with a suitable battery storage [23]. The maximum PVmodule rating in an SHS kit is expected not to exceed 350 Wp as per the current standards[24]. Figure 1.7 shows an example of an SHS from the company BBOXX active in EastAfrica; the SHS contains a PV module, battery storage with power electronics, and DCloads. The SHS is by definition standalone, i.e., it is not connected to the grid. The termsolar home systems may be used interchangeably with a standalone PV system, althoughthe term SHS has largely come to be used in the context of off-grid electrification.

Figure 1.7: Example of an SHS consisting of PV panel, battery storage and DC loads (model bPower50; imagecourtesy of BBOXX Ltd.).

In terms of the effectiveness of off-grid solar-based electrification at the householdlevel, a total of 124 million people have benefited from using off-grid solar between 2010and 2016 — 100 million people have used pico-solar products like solar lanterns (< 11Wp), and 24 million have used SHS [25].

1.2. LIMITATIONS OF SHSDespite being the natural technology choice to tackle electrification woes, there aremultiple reasons why SHS have not been able to achieve scale or fully eradicate energypoverty. This section details the inherent problems endemic to state-of-the-art SHS.

1

8 1. INTRODUCTION

1.2.1. COST

Even though the sun’s energy is freely available, the system for converting the free resourceinto useful electricity — SHS — incurs significant costs. Especially given the targetsegment for electrification, cost remains a stumbling block for SHS. As of 2016, while thePV module costs for SHS reached $1–2/Wp, the total initial system costs were significantlyhigher, ranging from around $4–15/Wp for different systems across Africa [26]. These costfigures vary greatly over time and across regions. However, in general, the initial costs ofSHS are typically more than 75% of the total lifetime costs, and these upfront costs can beequivalent to a low-income rural household’s earnings over a year [27].

While a direct impact on technology costs depends largely on the technology learningcurves, there are still other avenues where technical and technological interventions inthe context of SHS can yield direct dividends while alleviating cost-related woes. Theseother ‘avenues’ are also problems translatable to present-day SHS technical design, asexplained in Section 1.2.2 to Section 1.2.6.

1.2.2. BATTERY IN AN SHSThe battery is a vital component of the SHS, enabling energy storage of the PV output(output power generated by the PV modules), which can be utilized in the absence ofsunlight. However, the sizable proportion of upfront battery costs in total system costsmakes the battery the most expensive SHS component. This fact, coupled with the lowbattery lifetimes (sometimes even as low as 3 years [28]), makes battery-costs the mostrelevant in SHS design. Battery-costs recur not just in terms of the upfront costs, but alsoin terms of the replacements during SHS lifetime, thereby making battery lifetime a criticalparameter in SHS applications [29]. Additionally, accurate sizing of the battery can impactbattery lifetime [30], underlining the importance of battery lifetime as a design parameterto be taken into account while dimensioning an SHS, which also has implications on thetotal system costs. Therefore, the battery lifetime needs to be estimated based on theanticipated application in the context of SHS at the SHS design stage.

1.2.3. OPTIMAL SYSTEM SIZING

SHS dimensioning or sizing comprises the PV sizing (the rated output in Wp), batterysizing (the rated energy capacity), and the sizing of the power converters (rated power).PV sizing mainly depends on the total energy needed from the PV generator 3, which inturn depends on the load profile. A lower than adequate PV size results in system failureor a high number of loss of load events, i.e., loss of load probability (LLP). On the otherhand, a larger than adequate PV size results in wastage of energy along with higher systemcosts.

The battery is the most expensive SHS component while suffering from low lifetimesas compared to other SHS components. Additionally, a smaller than adequate batterysize will result in failure to meet the load requirements, while an oversized battery willdrastically increase the upfront costs of the system. Moreover, battery replacements areoften an additional hassle in the context of remote rural areas.

3The PV module is also be called as the PV generator, as in the context of SHS, the PV module is the only sourceof power generation.

1.2. LIMITATIONS OF SHS

1

9

For a given electricity demand, the optimal system size will reduce the wastage ofenergy in the form of excess power generation, increase the power supply availability, andincrease the estimated battery lifetime while keeping the upfront costs as low as possible.Therefore, the optimal sizing of SHS is a critical task for designing the SHS to cater tooff-grid electricity needs.

1.2.4. LOAD DEMAND IN OFF-GRID SYSTEMSLoad demand of a household can be quantified in the form of an electrical load profile. Aload profile can be defined as the power demand of an energy system mapped over time.A load profile not only captures the electricity demand of the user but also serves as a vitalinput to the electrical system design. Especially in the case of off-grid power systems likethe SHS, reliable apriori knowledge of the load profile is extremely helpful in the electricalsizing the system (i.e., deciding the PV rating and the battery capacity).

In fact, load profile, even if coarsely estimated, is almost always the starting point in anoff-grid, standalone PV system design [31]. Load profiles can have a profound impact onthe performance as well as design decisions in off-grid systems [32]. A better knowledgeof load profile makes for a more optimal off-grid electrical system design. Conversely, thelack of an appropriate load profile leads to either oversizing or undersizing the system,thereby causing an unhealthy trade-off between system costs and power availability [33].

Moreover, there is a lack of load profile data for off-grid users, especially when con-sidering electricity needs above basic lighting and mobile phone charging. Additionally,it is challenging to design an energy system for a user without an electricity footprint inthe past. Furthermore, even if the basic electricity demands are met, the electricity needsof a user in the off-grid context keep increasing with time [33, 34]. Lastly, the off-gridsector has seen the growth of the so-called super-efficient DC appliances in the last fewyears [35]. Therefore, in order to design and dimension off-grid SHS, load profiles areneeded that not only capture different levels of electricity usage, but also take into accountthe load consumption in line with the currently available super-efficient DC appliances.Hence, constructing load profiles for various levels of electrification would be one of thefirst problems to be tackled in this dissertation.

1.2.5. POWER AVAILABILITYPower availability is another limitation of SHS when compared to conventional grid exten-sion. While state-of-the-art SHS are extremely useful for powering a few DC appliancesand LED lights, they still face a power ceiling that prevents the users from using highpower appliances like pumps, washing machines, etc. The limited nature of the PV powerproduction, coupled with a limited capacity of the PV module and battery storage, meansthat the households in question are often restricted not just to the appliance choicebut also the duration of appliance usage. Again, this translates back to the challenge ofoptimal sizing, while also being able to accommodate a growing load profile.

1.2.6. CLIMBING UP THE ELECTRIFICATION LADDERAs mentioned in Section 1.2.4, the electricity needs of a household keep increasing withtime. This phenomenon can be visualized in the form of the so-called energy ladder, asreferred to in literature before [36]. Specifically in the context of electrification, it is also

1

10 1. INTRODUCTION

referred to as the electrification ladder [37, 33]. Figure 1.8 illustrates the concept of theelectrification ladder. At the lowest rung is an unelectrified household with no access toelectricity, represented by a fossil-based fuel source (e.g., kerosene lamp). Higher rungs ofthe ladder represent the increase in the number of appliances and therefore, the overallelectricity demand.

5 W

40 W

100 W

400 W

Figure 1.8: Concept illustration of the electrification ladder. As a household scales up the ladder, the electricitydemand increases through the addition of higher power (or greater number of) appliances relative to thepreceding rungs of the ladder. The power levels of the appliances are indicative only.

In their current form, present-day SHS are only able to fulfill the electricity needsin the lower rungs of the electrification ladder. Simplistic oversizing or over-stackingof PV and battery in an incumbent SHS to reach highest levels of electrification mightbe economically unviable. This is where the role of rural microgrids could be exploredto achieve higher levels of electrification, especially as a solution that goes further thanstandalone SHS yet without incurring the usual problems associated with conventionalgrid extension. This also opens up multiple avenues of exploration. For instance, thespecific benefits for going from standalone SHS to rural microgrids, and the optimalmicrogrid topology with decentralized power generation and storage while taking intoaccount the spatial spread of the households, are topics that can be suitably addressedthrough scientific investigation. This will be one of the topics to be explored in thisdissertation, as explained in Section 1.5. A comparison between SHS and microgrids aspotential electrification pathways is discussed in detail in Chapter 2.

Multi-tier framework for measuring household electricity access The different rungsof the conceptual electrification ladder can be alternatively viewed through the multi-tierframework (MTF) for measuring household electricity access [38], which describes 5distinct tiers of household electricity access. This is discussed in detail in Chapter 2. Whilethe electrification ladder is a useful concept to visualize the increasing electricity needs ofa household, the MTF gives a quantified categorization of the energy demand per tier ofelectricity access, as discussed later in Section 2.1.1. In this dissertation, the MTF will beused for the quantified distinction of different levels of electricity demand to be satisfiedthrough off-grid solar-based electrification.

1.3. SCOPE, OBJECTIVE AND RESEARCH QUESTIONS

1

11

1.3. SCOPE, OBJECTIVE AND RESEARCH QUESTIONS

1.3.1. SCOPE OF THIS DISSERTATIONAs discussed in Section 1.2, the state-of-the-art SHS is limited by several problems that arelargely inter-related. The greater issue of attaining SDG 7 and eradicating energy povertywarrants a multi-disciplinary approach spanning technology, policy, business, finance,user-centered design, amongst others. However, owing to the technical nature of thisdissertation, the scope of the work discussed in this dissertation is limited to the extentwhere technology can address the problems described in Section 1.2. Therefore, specialfocus is laid on the problems described in Section 1.2.2 to Section 1.2.6, without directlylooking at cost itself as a problem to be specifically solved. Keeping this in mind, theobjective and research questions for the work conducted in this dissertation are describedbelow.

1.3.2. MAIN OBJECTIVEThe main objective of this dissertation is stated as follows.

“To analyze the technological limits of Solar Home Systems (SHS) in terms of powerlevel, availability and energy storage for achieving universal electrification.”

1.3.3. RESEARCH QUESTIONSThe main objective can be broken down into the following research questions.

RQ1 What are the main (technical) limitations of the present electrification pathways inachieving universal electrification? Ch2

RQ2 How to construct load profiles for the various levels of off-grid electrification thatenable the design of off-grid solar-based solutions? Ch3

RQ3 How to estimate battery lifetime for a given SHS application at the SHS designstage? Ch4

RQ4 How to optimize the SHS size to satisfy the energy needs for various levels of elec-trification, while also ensuring maximum levels of power availability, minimizingexcess energy generation, and maximizing battery lifetime? Ch5

RQ5 What are the quantitative benefits on the battery storage sizing and power availabil-ity when going from standalone SHS to SHS-based microgrids that enable powersharing? Ch6

RQ6 How to find the optimal microgrid topology for interconnection of SHS in remoteoff-grid regions, taking into account the spatial spread of the households? Ch7

1.4. RESEARCH PUBLICATIONSThe research publications relevant to this dissertation are listed below along with thecorresponding chapters that are based on them.

1

12 1. INTRODUCTION

JOURNAL PUBLICATIONSJ1 N. Narayan, Z. Qin, J. Popovic-Gerber, J. C. Diehl, P. Bauer, M. Zeman. Stochastic

load profile construction for the multi-tier framework for household electricityaccess using off-grid DC appliances, Energy Efficiency, In Topical Collection ofJournal Energy Efficiency on Off-grid Appliances and Smart Controls for EnergyAccess, Springer 2018. (Ch 3)

J2 N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, M.Zeman. Estimating battery lifetimes in Solar Home System design using a practicalmodelling methodology, Applied Energy, Volume 228, 2018, Pages 1629-1639, ISSN0306-2619. (Ch 4)

J3 N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer,M. Zeman. Exploring the boundaries of Solar Home Systems (SHS) for off-gridelectrification: Optimal SHS sizing for the multi-tier framework for householdelectricity access. Applied Energy, 240, 2019, Pages 907-917. (Ch 5)

J4 N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, M.Zeman. Quantifying the Benefits of a Solar Home System-Based DC Microgrid forRural Electrification. Energies, 2019, 12(5), 938. (Ch 6)

J5 N. Narayan, M. Tagliapietra, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Zeman.Optimal microgrid layout using Geographic Information System and graph theoryconcepts, submitted. (Ch 7)

J6 N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Zeman.The long road to universal electrification: A critical look at present pathways andchallenges, submitted. (Ch 2)

J7 Y. Yu, N. Narayan, V. Vega-Garita, J. Popovic-Gerber, Z. Qin, M. Wagemaker, P. Bauer,M. Zeman. Constructing Accurate Equivalent Electrical Circuit Models of LithiumIron Phosphate and Lead–Acid Battery Cells for Solar Home System Applicationsin Energies, 2018 (special issue on Battery Storage Technology for a SustainableFuture), doi: https://doi.org/10.3390/en11092305. (Appendix A)

CONFERENCE PUBLICATIONSC1 N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P. Bauer and M. Ze-

man, "A simple methodology for estimating battery lifetimes in Solar Home Systemdesign," 2017 IEEE AFRICON, Cape Town, 2017, pp. 1195-1201. doi: 10.1109/AFR-CON.2017.8095652. (Ch 4)

C2 T. den Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer and M. Zeman, "Understanding the present and the future electricityneeds: Consequences for design of future Solar Home Systems for off-grid ruralelectrification," 2017 International Conference on the Domestic Use of Energy (DUE),Cape Town, 2017, pp. 8-15. doi: 10.23919/DUE.2017.7931816. (Partly Ch 2)

1.5. DISSERTATION LAYOUT

1

13

C3 N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer and M. Zeman, "Amodeling methodology to evaluate the impact of temperature on Solar Home Sys-tems for rural electrification," 2018 IEEE International Energy Conference (ENER-GYCON), Limassol, Cyprus, 2018, pp. 1-6. doi: 10.1109/ENERGYCON.2018.8398756(Appendix B)

C4 N. Narayan, B. O-Malik, L. Mackay, Z. Qin, J. Popovic-Gerber, P. Bauer and M.Zeman, "Decentralized Control-Scheme for DC-Interconnected Solar Home Systemsfor Rural Electrification," 2019 IEEE International Conference on DC Microgrids(ICDCM), Matsue, Japan, 2019, pp. 1-6. (Appendix C)

1.5. DISSERTATION LAYOUTFigure 1.9 illustrates the interdependency between the various chapters in the dissertation.Additional thematic separations have been identified, along with the research questionsthat will be answered and the research publications that the chapters are based on, asseen in Figure 1.9.

CHAPTER 1. INTRODUCTION

This chapter introduces the background, motivation, scope, objective, research questions,and layout of the dissertation.

CHAPTER 2. UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT PATHWAYS AND

CHALLENGES

This chapter discusses the 3 main electrification pathways, viz., grid extension, central-ized microgrids, and standalone solar-based solutions like pico-solar and SHS whileunderstanding their relative merits and demerits. Additionally, the main bottlenecks withSHS for large scale electrification and moving up the electrification ladder are discussed.This chapter sets the general directions for the technological explorations presented inthe rest of the dissertation.

CHAPTER 3. LOAD PROFILE CONSTRUCTION

This chapter presents the methodology followed to quantify the electricity demand of thehouseholds for various tiers of electricity access in terms of load profiles. In the context ofelectrification, reliable data for load profiles is often non-existent or limited to the lowesttiers of electricity access. Additionally, the off-grid appliances are rapidly evolving, withthe advent of the so-called super-efficient off-grid dc appliances, like TV, fan, etc. Thischapter details the construction of a stochastic load profile construction tool, resulting inrepresentative load profiles for various tiers of electricity consumption while consideringthe super-efficient off-grid dc appliances.

CHAPTER 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

Estimating battery lifetime is a critical task for SHS design. However, it is also a complextask due to the reliance on experimental data or modelling cell level electrochemicalphenomena for specific battery technologies and application use-case. This chapterpresents a practical, non-empirical battery lifetime estimation methodology specific

1

14 1. INTRODUCTION

1. Introduction

2. Reviewingelectrification

pathways

3. StochasticLoad profileconstruction

4. Estimatingbattery lifetimein Solar Home

Systems

5. OptimalSHS Sizingfor differentelectrifica-tion levels

6. Quantifyingbenefits ofSHS-basedmicrogrid

7. Optimalmicrogridtopologiesusing GIS

8. Conclusions

SH

S-M

icro

grid

Con

clus

ion

Bat

tery

and

SH

SLo

adpr

ofile

Intro

duct

ion

and

back

grou

nd

RQ1

RQ2

RQ3 RQ4

RQ5 RQ6

J6,C2

J1

J2,C1 J3

J4 J5

Figure 1.9: Dissertation layout showing the interdependencies between chapters and corresponding researchquestions.

to the application and the available candidate battery choices. An application-specificSHS simulation is carried out, and the battery activity is analyzed. A practical dynamic

1.5. DISSERTATION LAYOUT

1

15

battery lifetime estimation method is introduced, which captures the fading capacity ofthe battery dynamically through every micro-cycle.

CHAPTER 5. OPTIMAL SYSTEM SIZING FOR SHSThis chapter explores the possibilities of using SHS for climbing up the energy ladder. Anoptimal standalone system size is calculated for each tier of energy consumption, takinginto account the battery lifetime, temperature impact on SHS performance, power supplyavailability in terms of LLP, and excess PV energy. The optimal system sizing methodologyrelies on a multi-objective optimization using a genetic algorithm.

CHAPTER 6. QUANTIFYING BENEFITS OF INTERCONNECTED SHS-BASED DC MICROGRID

As standalone SHS are inadequate in meeting the electricity needs of higher tiers ofelectrification, an alternate solution is explored in this chapter with SHS having relativelylower battery sizes while sharing energy. A modular and scalable architecture for sucha bottom-up, interconnected SHS-based architecture is introduced, and the benefits ofthe microgrid over standalone SHS are quantified in terms of lower battery sizes and thedefined system metrics like LLP.

CHAPTER 7. OPTIMAL MICROGRID TOPOLOGIES USING GEO-INFORMATION SYSTEM (GIS)FOR MICROGRID PLANNING THROUGH SHS INTERCONNECTION

Optimal microgrid topologies are usually based on merely a comparison between astandard set of topologies, like ring, bus, spider, and radial topology. This chapter presentsa GIS-based methodology that looks at the geo-spatial spreads of households in remoteoff-grid settlements. The GIS-based household points obtained are then transformed in aplanar coordinate system so that graph theory-based concepts like Minimum SpanningTree can be applied. Thus, various topologies are considered apart from those existingin microgrid literature. The graph theory-based approach yields an interesting trade-offbetween total installed microgrid cabling needed to establish the interconnectivity andthe potential operational parameters like line losses and line congestion.

CHAPTER 8. CONCLUSIONS

This chapter contains the key conclusions and point-by-point answers to the researchquestions. Additionally, it explains the scientific implications of the research findings andits broader significance for the society. Finally, future work and recommendations arediscussed.

APPENDICES

Three appendices are presented in this dissertation.

• Appendix A Equivalent electrical circuit models of lithium iron phosphate andlead-acid battery cells for solar home system applications are presented in thisappendix chapter. This work is based on experiments performed at the cell level forthe two battery technologies.

• Appendix B A modeling methodology to evaluate the temperature impact on SHS(in terms of PV yield and battery lifetime) is discussed in this appendix chapter.

1

16 REFERENCES

• Appendix C A decentralized control scheme for interconnected solar home systemsfor dc microgrids is presented in this appendix chapter. Simulation results are alsodiscussed for a case study with 5 households exchanging power.

REFERENCES[1] IEA, Energy Access Outlook 2017 - From Poverty to Prosperity. Organization for

Economic Cooperation and Development, International Energy Agency, 1 ed., 2017.

[2] International Energy Agency, Energy Access Outlook 2017 - from poverty to prosperity.IEA, 2017.

[3] United Nations, “Sustainable development goals.” https://www.un.org/sustainabledevelopment/sustainable-development-goals/. Accessed:2017-05-19.

[4] R. Spalding-Fecher, “Health benefits of electrification in developing countries: aquantitative assessment in south africa,” Energy for Sustainable Development, vol. 9,no. 1, pp. 53–62, 2005.

[5] S. Ray, B. Ghosh, S. Bardhan, and B. Bhattacharyya, “Studies on the impact of energyquality on human development index,” Renewable energy, vol. 92, pp. 117–126, 2016.

[6] B. Pillot, M. Muselli, P. Poggi, and J. B. Dias, “Historical trends in global energy policyand renewable power system issues in sub-saharan africa: The case of solar pv,”Energy policy, vol. 127, pp. 113–124, 2019.

[7] J. Sušnik and P. van der Zaag, “Correlation and causation between the un humandevelopment index and national and personal wealth and resource exploitation,”Economic Research-Ekonomska Istraživanja, vol. 30, no. 1, pp. 1705–1723, 2017.

[8] E.-j. Quak, “Lighting and electricity services for off-grid populations in sub-saharaafrica,” Knowledge, evidence and learning for development, Tech. Rep., 2018.

[9] Global LEAP, “The State of the Off-Grid Appliance Market,” tech. rep., Global LEAPLighting and Energy Access Partnership, 2016.

[10] S. Buluswar, Z. Friedman, P. Mehta, S. Mitra, and R. Sathre, 50 Breakthroughs -Critical scientific and technological advances needed for sustainable global develop-ment. LIGTT, Institute for Globally Transformative Technologies, Lawrence BerkeleyNational Lab, 2014.

[11] ESMAP, “State of Electricity Access Report 2017,” tech. rep., World Bank Group,ESMAP and SE4ALL, 2017.

[12] D. F. Barnes and H. Samad, Measuring the Benefits of Energy Access: A Handbook forDevelopment Practitioners. Inter-American Development Bank, 2018.

[13] Lighting Global, Bloomberg New Energy Finance and GOGLA, “Off-grid solar markettrends report 2016,” Bloomberg New Energy Finance and Lighting Global in coopera-tion with the Global Off-Grid Lighting Association (GOGLA), 2016.

REFERENCES

1

17

[14] E. F. Merchant, “Irena: Global renewable energy prices will be competitive withfossil fuels by 2020.” https://www.greentechmedia.com/articles/read/irena-renewable-energy-competitive-fossil-fuels-2020#gs.3y8akc.Accessed: 2019-02-30.

[15] IRENA, “Renewable Power Generation Costs in 2017,” International Renewable En-ergy Agency, Tech. Rep., 2018.

[16] I. I. U. World Bank, and World Health Organization, “Tracking SDG7: The energyprogress report 2018,” tech. rep., International Energy Agency, International Re-newable Energy Agency, United Nations Statistics Division, World Bank, and WorldHealth Organization, 2018.

[17] D. Palit and K. R. Bandyopadhyay, “Rural electricity access in south asia: Is gridextension the remedy? a critical review,” Renewable and Sustainable Energy Reviews,vol. 60, pp. 1505–1515, 2016.

[18] T. Urmee, D. Harries, and A. Schlapfer, “Issues related to rural electrification usingrenewable energy in developing countries of asia and pacific,” Renewable Energy,vol. 34, no. 2, pp. 354–357, 2009.

[19] N. Narayan, J. Popovic, J. C. Diehl, S. Silvester, P. Bauer, and M. Zeman, “Developingfor developing nations: Exploring an affordable solar home system design,” in 2016IEEE Global Humanitarian Technology Conference (GHTC), pp. 474–480, Oct 2016.

[20] N. Otuki, “Half of kenya power clients use sh10daily.” https://www.businessdailyafrica.com/economy/Half-of-Kenya-Power-clients-use-Sh10-daily/3946234-4643380-tmq6js/index.html. Accessed: 2019-01-28.

[21] Gabriel Davies, Matt Tilleard and Lucy Shaw, “Private mini-grid firms deserve achance to compete against slow utilities in africa.” https://www.greentechmedia.com/articles/read/a-faster-path-to-rural-electrification#gs.4p99x3. Accessed: 2017-05-19.

[22] IEA, World Energy Outlook 2016. Organization for Economic Cooperation and Devel-opment, International Energy Agency, 1 ed., 2016.

[23] Global Off-Grid Lighting Association, “Global off-grid solar market report - semi-annual sales and impact data,” tech. rep., GOGLA, Lighting Global and Berenschot,2018.

[24] Lighting Global, “Solar home system kit quality standards version 2.5,” LightingGlobal, Tech. Rep., December 2018.

[25] IRENA, “Off-grid renewable energy solutions to expand electricity access: An op-portunity not to be missed,” International Renewable Energy Agency, Tech. Rep.,2019.

1

18 REFERENCES

[26] IRENA, “Solar PV in Africa: Costs and Markets,” International Renewable EnergyAgency, Tech. Rep., 2016.

[27] T. Urmee, D. Harries, and H.-G. Holtorf, Overview of Financing Mechanisms for SolarHome Systems in Developing Countries, pp. 49–77. Cham: Springer InternationalPublishing, 2016.

[28] Good Solar Initiative, Quality Charter: Technical and Service Quality Standards forAccredited Solar Suppliers. SNV Netherlands Development Organisation, 2016.

[29] N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P. Bauer, and M. Zeman,“A simple methodology for estimating battery lifetimes in solar home system design,”in 2017 IEEE AFRICON, pp. 1195–1201, Sept 2017.

[30] N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Estimating battery lifetimes in solar home system design using a practi-cal modelling methodology,” Applied Energy, vol. 228, pp. 1629 – 1639, 2018.

[31] A. H. Smets, K. Jäger, O. Isabella, R. Van Swaaij, and M. Zeman, Solar Energy, thePhysics and Engineering of Photovoltaic Conversion Technologies and Systems. UITCambridge, 2016.

[32] S. Treado, “The effect of electric load profiles on the performance of off-grid resi-dential hybrid renewable energy systems,” Energies, vol. 8, no. 10, pp. 11120–11138,2015.

[33] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrifica-tion,” in 2017 International Conference on the Domestic Use of Energy (DUE), pp. 8–15,April 2017.

[34] M. Gustavsson, “With time comes increased loads: An analysis of solar home systemuse in lundazi, zambia,” Renewable Energy, vol. 32, no. 5, pp. 796–813, 2007.

[35] N. Narayan, Z. Qin, J. Popovic-Gerber, J.-C. Diehl, P. Bauer, and M. Zeman, “Stochasticload profile construction for the multi-tier framework for household electricityaccess using off-grid dc appliances,” Energy Efficiency, Nov 2018.

[36] I. Bisaga and P. Parikh, “To climb or not to climb? investigating energy use behaviouramong solar home system adopters through energy ladder and social practice lens,”Energy research & social science, vol. 44, pp. 293–303, 2018.

[37] B. Tenenbaum, C. Greacen, T. Siyambalapitiya, and J. Knuckles, From the bottom up:how small power producers and mini-grids can deliver electrification and renewableenergy in Africa. World Bank Publications, 2014.

[38] M. Bhatia and N. Angelou, Beyond Connection, Energy Access Redefined. The WorldBank, 2015.

2THE LONG ROAD TO UNIVERSAL

ELECTRIFICATION: A CRITICAL

LOOK AT PRESENT PATHWAYS AND

CHALLENGES

We will make electricity so cheap that only the rich will burn candles

Thomas A. Edison

ABSTRACTIn this chapter, the three different electrification pathways — grid extension, centralizedmicrogrids, and standalone solar-based solutions like pico-solar and SHS — are criticallyexamined while understanding their relative merits and demerits. Grid extension canprovide broad-scale access at low LCOE values but requires a certain electricity demandthreshold and population density to justify investments. To a lesser extent, centralized(off-grid) microgrids also require a minimum demand threshold and knowledge of theelectricity demand. Solar-based solutions are the main focus in terms of off-grid elec-trification in this chapter, in line with the scope of this dissertation, as mentioned in

This chapter is based on the following publications:

1. Narayan, N., Vega-Garita, V., Qin, Z., Popovic-Gerber, J., Bauer, P., & Zeman, M. The long road to universalelectrification: A critical look at present pathways and challenges, submitted.

2. (Partly based) T. den Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber, P.Bauer and M. Zeman. Understanding the present and the future electricity needs: Consequences fordesign of future Solar Home Systems for off-grid rural electrification, 2017 International Conference onthe Domestic Use of Energy (DUE), Cape Town, 2017, pp. 8-15.

19

2

202. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

Chapter 1. In recent times, decentralized solar-based off-grid solutions like pico-solarand SHS have shown the highest adoption rates and promising impetus with respectto basic lighting and electricity for powering small appliances. However, the burningquestion is — from lighting a million to empowering a billion – can solar home systems getus there? The two main roadblocks for SHS are discussed, and the requirements from theideal electrification pathway are introduced. A bottom-up, interconnected SHS-basedelectrification pathway is proposed that can form the missing link among the presentelectrification pathways.

OUTLINEThis chapter contains 7 sections. Section 2.1 introduces the problem of universal elec-trification with the 3 electrification pathways presently in use. Sections 2.2–2.4 discusseach of the electrification pathways in detail. Section 2.5 presents the significant chal-lenges hindering the state-of-the-art SHS from becoming the electrification pathway ofchoice in the long run. Section 2.6 describes the requirements from an ideal pathwaythat can overcome the current challenges and discusses the comparative qualities of eachelectrification pathway. Section 2.7 the challenges with the SHS-based vision towardsuniversal electrification. Finally, Section 2.8 concludes the chapter with closing thoughtson universal electrification.

2.1. INTRODUCTIONUniversal electrification is a monumental challenge facing humankind today. Just undera billion people globally lacked access to any form of electricity at the end of 2017 [1],an (inadequate) improvement since 2000, when there were 1.7 billion people withoutelectricity [2]. Three main solutions have been tried, with varying degrees of success, inthe last decade to increase the umbrella of electrification. These are:

1. Grid extension

2. Off-grid micro (or mini-) grids

3. Standalone household level solutions like pico-solar and solar home systems (SHS)

While solar-based standalone solutions have gained prominence in the last few years,the overall electrification efforts have not been rapid enough. The UN also took note ofthe progress on SDG 7, saying it “remains too slow to be on track to meet the global energytargets for 2030” [3].

2.1.1. MULTI-TIER FRAMEWORK FOR MEASURING ELECTRICITY ACCESSBefore proceeding to understand the electrification pathways, it is first essential to under-stand the concept and significance of the multi-tier framework for measuring householdelectricity access.

In the past, governments have often defined electricity access as a connection to anelectrical grid, whether or not that connection is reliable. Additionally, large sections ofthe people (e.g., a whole village) have also been called electrified if a small percentageof the households is electrified [4]. Until the previous decade, electricity access was

2.2. PATHWAY 1: GRID EXTENSION

2

21

Table 2.1: Multi-tier matrix for measuring access to household electricity supply. Sourced from [5].

Tier 1 Tier 2 Tier 3 Tier 4 Tier 5

Energy and peakpower rating

> 12 Wh& > 3 W

> 200 Wh& > 50 W

> 1 kWh& > 200 W

> 3.4 kWh & > 800 W > 8.2 kWh & > 2 kW

Availability(hrs/day)

> 4 > 4 > 8 > 16 > 23

Availability(hrs/evening)

> 1 > 2 > 3 > 4 > 4

Reliability — — — < 14 disruptions per week < 3 disruptions per weekQuality — — — Voltage problems do not affect the use of desired appliancesAffordability — — — Cost of 365 kWh/year < 5% of household incomeLegality — — — Bill is paid to the utility or authorized representativeHealth &Safety

— — — Absence of past accidents and high risk perception in the future

largely looked at as a have or have-not condition. However, such oversimplification isa dangerously narrow view of understanding and acting towards the electricity accessproblem. Such binary metrics were therefore considered insufficient, and a multi-tierframework (MTF) was proposed in [5], which captures the multi-dimensional nature ofelectricity access. Table 2.1 presents the MTF as described in [5].

In terms of the electrification ladder mentioned previously in Chapter 1 in Figure 1.8,the climb up the ladder can now be alternatively seen as the movement across the tiers ofthe MTF.

SCOPE OF THIS CHAPTER

In this chapter, the three main electrification pathways currently in use are analyzedwhile discussing the relative merits and demerits for each pathway. Special attention ispaid to the solar-based off-grid electrification. This is because off-grid renewable energy-based electrification has emerged in the last decade as a mainstream, cost-competitivealternative to grid-based electricity. For instance, between 2011 and 2016 alone, morethan 133 million people benefited from off-grid renewables [6]. We critically look at someof the most pressing problems that off-grid solar-based electrification is/will be facing inachieving scale en route to universal electrification.

Furthermore, it must be noted that technology is only one part of the larger puzzleof electrification. Business and policy aspects also need significant attention, but thescope of this chapter is largely limited to the technical and technological considerationsof electrification.

2.2. PATHWAY 1: GRID EXTENSIONGrid-based electricity refers to a household having its electricity derived from a localnetwork that is connected to the larger transmission network, where the grid is typicallypowered by large centralized power plants. Grid extension refers to the act of expandingthe national grid’s network to provide electricity access. Grid extension includes installingmedium voltage distribution lines as well as adding new connections to the householdsneeding electricity access [7].

Grid extension has historically been the most common way to provide electricityaccess to un(der-)electrified regions and communities. Apart from being a common

2

222. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

approach, it can also be the most cost-effective way to provide electricity access at veryhigh levels of electricity demand, e.g., for tier 4 and tier 5 electricity access.

However, grid extension is cost-effective only if the target region has a certain popula-tion density and energy demand. Otherwise, the extension efforts are almost certain toincur losses. For example, 55% of all customers serviced by Kenya Power and LightingCorporation (KLPC) spend less than $3 a month on electricity, pushing the payback pe-riod of a typical KLPC grid connection to 44 years, even for higher levels of consumption[8]. The high investment costs for grid extension have been estimated in the order ofe22750/km for transmission line ande12000/km for distribution line in most Africancountries; in comparison, the grid-based retail electricity tariffs in these countries couldrange frome0.04/kWh (subsidized) to overe0.23/kWh (non-subsidized) [9]. Inadequaterevenues also result in progressive deterioration of the reliability of the transmissionand distribution network. Unsurprisingly, the compounding effects of these inadequaterevenues led to 81% of sub-Saharan African utilities reporting net financial losses in 2013[10]. Unfortunately, the present reality is that conventional African utilities tend to run aloss every time they connect to a rural customer [8].

While the lack of affordability by the consumers can hamper the viability of thenational grid as an electrification option, there are also examples where the infrastructurelags the people’s ability to pay. There are several countries where people above the povertyline, who could potentially afford to pay for electricity but still do not have access dueto infrastructural issues. For example, it was estimated that in sub-Saharan Africa, 120million people were living above the poverty line without access to electricity [2]. Otherfactors like topographical issues, regulatory hurdles, and sparse populations can alsomake the investment and maintenance of grid extension less favourable compared tooff-grid solutions [2].

Figure 2.1: So close, yet so far: A rural household in Aligarh, Uttar Pradesh (India) with installed SHS. Grid lineshave passed close to the house for years, yet the household was without any grid-based electricity until 2017.Photo credit: Gaurav Manchanda.

Fieldwork carried out in 2017 related to this PhD project also highlighted the complex

2.2. PATHWAY 1: GRID EXTENSION

2

23

issues at the grass-roots level related to the limitations of national grid outreach [11].Figure 2.1 depicts a “powerful” image; it captures the harsh reality of electricity accessfor a rural household in a state in Northern India. A single solar panel can be seen on therooftop that also has cow dung cakes drying in the sun, which will be used as cookingfuel once dried. Making up the powerful irony in the picture are the grid lines passingtantalizingly close to the house, which had no grid-based electricity access until 2017despite the presence of the grid in the region for several years.

CHICKEN AND EGG PROBLEM?Grid-based electrification in low-income countries is often seen as a chicken and eggproblem. Having multiple areas for development to be pursued with inadequate budgets,low-income countries often do not have the means to invest in large-scale grid electrifica-tion. On the other hand, lack of electrification, as discussed before in Section 1.1.1, keepsthe progress towards development stymied.

2.2.1. COMPETITION WITH OFF-GRID RENEWABLESA significant factor that has led to grid expansion losing out on the cost-effectivenessfront for rapid electrification is the rise of the off-grid solutions. In particular, solar-basedoff-grid solutions like pico-solar SHS have enjoyed tremendous success in recent years asthe electrification solution of choice. This has mainly been due to the massive drop inprices of most pico-solar and SHS components. For example, PV price has fallen by morethan 80% from 2009 to 2017; LED lights and Li-ion batteries have dropped by 80% and 73%respectively between 2010 and 2016 [6]. Moreover, national grids often heavily depend onfossil fuels for power generation and are therefore counterproductive to progress on otherSDGs like SDG 13 (climate action) and SDG 11 (sustainable cities and communities).

However, it is usually a complex task to declare grid extension better or worse thanan off-grid alternative. Investigating the viability of grid extension for a particular targetlocation must take into account not just its distance from the existing grid, but also theexpected energy consumption and relative off-grid standalone system costs. Figure 2.2simplistically illustrates the conceptual cost curve as a function of distance assuming acertain energy demand and fixed off-grid system cost. The point where the grid extensioncosts and off-grid costs meet forms the breakeven distance for that particular case. To theleft of that distance, the grid-extension option is more economically favourable, while anoff-grid solution is favourable to the right of that breakeven distance.

Figure 2.3 depicts a conceptual comparison of breakeven annual electricity consump-tion as a function of upfront SHS system costs for a given target region. It is assumedthat different target regions would have different mean distances from the grid leadingto different curves. The annual electricity consumption as outlined by the MTF is alsomarked for the various tiers on this conceptual graph. Grid extension will be a cheaperoption if the annual consumption of a household in the target region is more than thebreakeven electricity level given by the curve. Although a conceptual illustration, it can beseen that the capital SHS costs need to be much lower to be able to justify a standalonesystem for higher tiers of consumption.

Therefore, it can be said that the precise determination of the viability of grid extensionfor the electrification of a particular target location depends on multiple factors as seen in

2

242. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

Co

st

Distance

off-grid system

grid extension

grid extension cheaper off-grid cheaper

Figure 2.2: Conceptual comparison of off-grid system and grid-extension cost-curves as a function of distancefrom the existing grid (adapted from [7]). Off-grid system costs are independent of distance.

Figure 2.2 and Figure 2.3. Furthermore, other non-technical factors like policy, regulations,and topography could pose additional challenges for grid extension.

0 3 6 9 12 150

1,000

2,000

3,000 tier 5

tier 4

tier 3

tier 2 tier 1

Total SHS cost [$/Wp]

Bre

akev

enen

ergy

con

sum

pti

on

[kW

h/y

ear/

ho

use

ho

ld]

Region 1Region 2

Figure 2.3: Illustrative breakeven annual consumption as a function of SHS upfront costs for two differentregions (adapted from [12]). The different tier consumptions of the MTF are also marked for reference. Tier 1 isnot clearly visible given the y-axis scale, as the annual energy consumption for tier 1 level is merely 4.38 kWh.

2.3. PATHWAY 2: (OFF-GRID) CENTRALIZED MICROGRIDS AND MINI-GRIDS

2

25

2.3. PATHWAY 2: (OFF-GRID) CENTRALIZED MICROGRIDS AND

MINI-GRIDS

2.3.1. CLARIFICATION IN TERMINOLOGYWhen it comes to off-grid electrification, there are no universally accepted definitionsfor, or distinctions between, mini-grids and microgrids. Several different definitions exist.For example, the IEA has defined mini-grids as "small grid systems linking a number ofhouseholds and other consumers" [13]. Although not specified in the context of off-gridelectrification, the National Renewable Energy Laboratory (NREL) defined microgrids as“a group of interconnected loads and distributed energy resources that acts as a singlecontrollable entity with respect to the grid” [14]. Mini-grids have also been defined assmall-scale distribution networks producing electricity from small generators, typicallyoperating below 11 kV, and providing power to one or more local communities [4]. In thecontext of electrification, generally, mini-grids and microgrids operate isolated from thegrid. Hence, this electrification pathway is referred to as “off-grid".

Some definitions also take into account power levels, although these limits may alsovary from one definition to another. For example, the Alliance for Rural Electrification(ARE) defines mini-grids as having capacities ranging from 10 kW to 10 MW, and micro-grids as being similar to mini-grids with generation capacities of 1–10 kW. In general,mini-grids and microgrids are standalone electrical power systems that serve multipleconsumers through wired connections, powered by PV modules, diesel gensets, or/andwind turbines [7]. A microgrid has a connotation of being smaller in power capacity, asalso seen in some of the power capacity-based definitions above.

Note: For the remainder of this dissertation, the term “microgrids" will be consistentlyused in the context of off-grid electrification. However, many of the arguments andideas discussed could also be interchangeably applied also to mini-grids, especially whencentralized power generation is considered.

2.3.2. CENTRALIZED MICROGRIDSMicrogrids for off-grid electrification can further be classified based on the power gen-eration technology used. For example, solar PV, biomass, micro-hydro, wind, diesel, etc.Hybrid microgrids are also possible, which use two or more power generation technolo-gies. Energy storage technologies like batteries are frequently used in microgrid projectsto improve the availability of power supply. The most common storage technologies arelead-acid and lithium-ion batteries. Sometimes, diesel is also used to replace the batteryin a hybrid microgrid, although this option incurs running fuel costs. Usually, the powergeneration is centralized, i.e., the power generation occurs centrally, and there is a distri-bution network responsible for supplying power to the consumers. In this dissertation,the main focus is on solar PV as the power generation technology of choice, owing to itshistorically low prices, modular nature, and ease of installation.

An example of a centralized microgrid architecture is shown in Figure 2.4. The termcentralized refers to the central power generation and storage, which is shown in Figure 2.4in the form of solar PV and battery storage. Additionally, given the recent proliferation ofthe so-called super-efficient DC appliances (as discussed later in Chapter 3), the microgrid

2

262. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

depicted in Figure 2.4 assumes a DC architecture.

Figure 2.4: An illustration of a centralized (DC) microgrid architecture. The term centralized refers to the centralpower generation (solar PV) and storage (battery). Each house is assumed to have a set of (DC) loads.

CENTRALIZED VS. DECENTRALIZED

Microgrids-based off-grid electrification is often referred to as being decentralized. It mustbe cautiously noted that the classification of these systems being decentralized stemsfrom collectively viewing the myriads of such microgrids juxtaposed with the centralnational grid. In that specific context, these microgrid systems can be together viewed asautonomous, and hence a decentralized way of delivering electricity. However, in termsof architecture, rural microgrids are usually centralized with central power generationand storage. Therefore, these off-grid microgrids have been classified as centralizedmicrogrids in this dissertation.

2.3.3. MICROGRIDS VS. GRID EXTENSIONWhere national electricity grids fail to reach, microgrids can step in with several benefits.Microgrids score over grid extension over several categories, as described below.

1. Transmission and connection costs. One of the biggest gains for microgrids incomparison to grid extension comes in the form of minimizing transmission costs,which can be up to 40% of the total grid-based electricity costs [15]. On the otherhand, as the microgrids usually serve areas within 5 km of the installation, theyincur minimal transmission losses compared to the grid. Typical mini-grid retailtariffs in Africa have seen to range from e0.1/kWh to e1.2/kWh, depending onlocal conditions and operators [9]. Average per connection capital cost of startinga microgrid has been estimated to start from as low as $50, as compared to theaverage per connection costs of grid extension, which have been estimated at $500[15].

2. Empowering local population. Furthermore, traditional national grids are largelya vertically integrated and regulated monopoly. On the other hand, rural microgrids

2.3. PATHWAY 2: (OFF-GRID) CENTRALIZED MICROGRIDS AND MINI-GRIDS

2

27

can be highly empowering to the local consumer population, not just in terms ofthe benefits of electricity itself, but in terms of local entrepreneurship opportunities.These microgrids also have the potential to improve gender equality through greaterinvolvement of women in the value chain [16].

3. Reliability of power supply. In terms of reliability, microgrids can often performbetter than ailing grids in urban and peri-urban areas of low- and middle-incomecountries. For example, private sector microgrids have reportedly observed 98%up-time in Tanzania as compared to the national grid’s reported usable voltagerange availability of 47% in Dar es Salaam [8].

4. Integrating renewables. In terms of generating power technology, microgrids areperfectly poised to utilize renewables, as opposed to grids. Since most of themicrogrids are to be installed from scratch, opting for renewable sources of energylike solar PV is much more economical in both capital costs as well as fuel costs.This way, the route to achieving SDG 7 can be without adversely affecting the overallclimate targets and hampering the progress towards other SDGs.

Furthermore, while there are many renewable energy technologies available, solarPV is the most viable and widely suited, as there is abundant sunshine in theelectricity-starved regions most of Asia and Africa, except for a few locations dueto excessive rains. In comparison, other renewable energy sources like (micro-)hydroelectric power have to depend on site-specific resources like a perennialstream.

The efficacy of (rural) microgrids in electrification has been established beyond doubt.The undeniable potential, as well as the necessity of rural microgrids, has been mentionedin several studies on future scenarios. In one of its scenarios, the IEA estimates 140 millionpeople in Africa to receive electricity access through mini-grids by 2040, necessitating thedevelopment of between 100,000 to 200,000 mini-grids [13].

2.3.4. MICROGRIDS VS. STANDALONE SHSFigure 2.5 shows a comparative plot of the total installed capacity of off-grid solar energysources for the years 2010 and 2016. A phenomenal growth in capacity can be seen foroff-grid solar, underlining a positive trend and a momentum that is poised to continue.Figure 2.6 shows a similar statistic for the number of people gaining access to electricitythrough off-grid solar between 2010 and 2016. The remarkable fact is that despite amuch smaller number proportion of people affected by the microgrids (nearly 124 millionthrough standalone PV as compared to 2.2 million tier 1 and 2 level PV microgrid and9 million including hybrid microgrids in 2016 [17]), the total installed capacity of PVmicrogrids is much more sizable at around 296 MW than SHS at around 215 MW andpico-solar at 58 MW, as seen in Figure 2.5. This can is attributed to the fact that theinstalled capacity of the microgrids, which are usually at a community scale, can be muchhigher than SHS and pico-solar, which are limited to the household level.

Compared to standalone systems like SHS, microgrids offer multiple advantages, asdiscussed below.

2

282. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

0 200 400 600 800 1,000 1,200

2010

2016

Total installed capacity (MW)

pico-solar SHS (11–50 Wp) SHS (>50 Wp) PV microgrid solar pumps other solar

Figure 2.5: Capacity of different installed off-grid solar energy sources including SHS, pico-solar, solar PVminigrids, solar pumps, and other uncategorized off-grid solar PV applications (data from [17]).

0 20 40 60 80 100 120

2010

2016

Total people connected (millions)

pico-solar SHS (11–50 Wp) SHS (>50 Wp) PV microgrid

Figure 2.6: Number of people with improved electricity access to off-grid solar energy sources including SHS,pico-solar and solar PV minigrids (data from [17]).

1. High power loads. Higher installed power capacity of the microgrids by definitionallows for higher power loads. Thus, people can potentially enjoy higher tiers ofelectricity access, as opposed to smaller standalone systems like SHS or pico-solarproducts, which often tend to keep the consumers locked in with respect to thechoice of appliances.

2. Load variance. Higher load variance can be achieved through microgrids as mul-tiple households are connected [18]. This leads to the added benefit of betterutilization of the installed resources like PV and battery, provided the given levels ofload consumption are commensurate with the installed power capacity.

3. Productive use of energy (PUE). Productive use of energy implies the use of energyfor supplementing income generating activities [19]. Usually, PUE appliances likepower tools, pumps, etc. are of higher power rating than basic consumptive house-hold appliances like fans, TV, and a small fridge [20]. These are better supported by

2.3. PATHWAY 2: (OFF-GRID) CENTRALIZED MICROGRIDS AND MINI-GRIDS

2

29

microgrids than SHS. More on PUE is discussed in Section 3.1.2.

4. Community buildings and loads. Microgrids are also suited for community build-ings like schools and hospitals, where a single SHS will not suffice. Additionally,community level loads can be met by microgrids more easily than SHS.

5. Cheaper unit system costs. Compared to SHS, microgrids can fundamentallyattain cheaper unit system costs due to economies of scale. However, the overallsystem cost can pose a significant challenge in the beginning.

2.3.5. DISADVANTAGES OF CENTRALIZED MICROGRIDSMicrogrids also come with a distinct set of disadvantages that are holding back thiselectrification pathway from achieving its full potential. These demerits are describedbelow.

1. High capital expenditures(CAPEX). High CAPEX costs are a major barrier leadingto slow uptake of microgrids for electrification. Unlike biomass plants, whichhave low CAPEX and high operational expenditures (OPEX), PV microgrids havehigh CAPEX and low OPEX. High CAPEX can be an impediment, especially if themicrogrid is operating on a private or community-owned model, as opposed to autility-owned model.

2. Arrival of the national grid. The arrival of the national grid has complicated con-sequences. On the one hand, the advent of the grid could be a perfect ending to amicrogrid story as the microgrid could (potentially) seamlessly integrate with thegrid. On the other hand, it could also bring about technical as well as financialconsequences. As a thumb rule, grid tariffs are usually lower than microgrid tariffsdue to cross-subsidization as well as economies of scale [9]. Additionally, certaintechnical standards need to be met before a microgrid could connect with the maingrid, e.g., voltage and frequency regulation, and islanding capabilities. Microgridscan turn into stranded assets if the arrival of the grid is not planned for. Regulatoryframeworks need to exist from the start if the integration of microgrids with thenational grid is desired.

3. Need for demand threshold. Similar to national grids, albeit, at a lower level,microgrids need a certain demand threshold to justify their initial investment.

4. Operational involvement with consumers. Although the OPEX costs are relativelylow in PV microgrids, high operational involvement is needed from the microgridoperators in terms of interaction with hundreds or thousands of households forrevenue collection. This part is unique to rural microgrids, as opposed to largesolar farms where the plant developers have minimal operational involvement. Theparticipation of well-established players has therefore been negligible in terms ofrural microgrids as opposed to smaller private entities.

5. Overall demand estimation. The peak power rating for the central installation ishinged on the demand estimation of (usually) hitherto un(der-)electrified com-munity. In some cases, the installation might even be oversized to account for the

2

302. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

future demand of the microgrid. Lower than expected consumption patterns couldlead to a risk of inadequate revenues, while higher than expected consumptioncould lead to potential blackouts.

2.4. PATHWAY 3: SOLAR-BASED STANDALONE SYSTEMSThis electrification pathway includes solar-based standalone systems like pico-solar(solar lanterns) and SHS. As discussed in Chapter 1, SHS are standalone PV systems rated11 Wp and above, usually not more than 350 Wp and mostly less than 100 Wp. Solarlanterns or pico-solar products are defined to have rated PV power of less than 11 Wp [21].Sometimes, further categorization of SHS and pico-solar exists, based on either wattage(for SHS), like 11–21 Wp, 21–50 Wp, 50–100 Wp, and > 100 Wp, or based on servicesprovided (for pico-solar), like single light, single light + mobile charging, multiple lights +mobile charging [22]. In any case, it is expected that pico-solar products can enable tier 1access, while SHS can enable at least tier 2 access.

(a) A solar LED lantern (d.light S30) with0.3 Wp solar module and integrated 60lm LED [23]

(b) A solar home system (d.light X850) with a 40 Wp panel and DCappliances [24]

Figure 2.7: Examples of a pico-solar product (solar lantern) and an SHS from the company d.light. Imagesobtained from [21].

Figure 2.7 shows examples of pico-solar and SHS products currently being usedin the off-grid market, helping to accelerate (decentralized) electrification. Figure 2.8shows a symbolic representation of SHS in a household along with a schematic blockdiagram representation. The SHS consists of a PV module, a suitably sized battery, powerelectronics for power conversion, and a set of DC loads.

Since pico-solar is limited to tier 1 access only, SHS is a more viable alternative fordecentralized electrification, and therefore, discussed in greater detail in the followingsections. Installation of SHS has improved the lives of off-grid consumers in multipleways, e.g., creating job opportunities, enhancing business revenue, unlocking additionalwork hours, and increasing monthly income, among others [25].

2.4.1. SHS VS. GRID AND MICROGRIDSPresent day SHS enjoys several advantages as the pathway of choice in acceleratingtowards last-mile electrification targets. Some of these advantages are discussed below.

2.4. PATHWAY 3: SOLAR-BASED STANDALONE SYSTEMS

2

31

(a) SHS symbol

PVpanel

Powerelectronics

Battery

DC loads

(b) Block diagram of an SHS

Figure 2.8: Symbolic and block diagram representations of an SHS.

1. Least cost option. Even though SHS score poorly in terms of LCOE ($/kWh) or unitsystem costs ($/Wp), they are by definition able to offer least total system costsby catering to a single household, as opposed to centralized microgrids and grids,which need to attain a certain scale of consumers to justify lower costs. SHS are aneconomically more appealing choice when facing conditions like long distancesfrom the grid, low density of population, and low electricity demands. Of course, theaffordability of the user depends largely on the target regions and specific financingmechanisms available, which is a different topic altogether, and not within thepurview of this dissertation.

2. Last mile connectivity. Historically, grid-based electricity has taken the longestto reach the remotest corners of a country. In terms of last mile connectivity,while rural microgrids perform better than national grids, SHS perform even betterthan microgrids as the electricity demand threshold is the lowest for SHS. Often,governments themselves subsidize the uptake of SHS to augment the nationalelectrification efforts.

3. Use of DC appliances. Many SHS providers are opting for a full DC system to saveon conversion losses. BBOXX, a company active in East Africa, supplies its SHSconsumers with its in-house portfolio of DC appliances. In Cambodia, the use ofAC appliances and inverters is discouraged, and end users are also advised to useefficient DC appliances with their SHS [26]. These so-called super-efficient DCappliances can significantly reduce the strain on SHS size for delivering the sameamount of electricity, thereby shifting the total cost distribution of an SHS alongwith resulting in cost savings [27, 2]. Additionally, the extent of the increasinglyefficient DC options available in the market today, as catalogued extensively in[28, 19], proves further the impetus that DC-based SHS enjoy going forward. Moreon these efficient appliances can be found in Chapter 3.

4. Better service through technology. The use of technology like data acquisitionand remote monitoring can significantly impact consumer satisfaction and uptake,which is otherwise a usual pain-point for SHS. Internet of things (IoT) is increasinglybeing used for implementing smart SHS to facilitate easy data acquisition and

2

322. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

analysis. For example, BBOXX, an SHS company operational in Rwanda, utilizedits SMART Solar platform in nearly 20000 SHS to better understand the user needs,enabling them to create scalable business models for energy access [29]. Thisleads to not just better user insights but also faster service and higher SHS usersatisfaction.

5. Rapid growth and adoption. Scaling up is often easier for SHS as opposed to mi-crogrids, because microgrids are often a customized solution from one installationto another. On the other hand, an SHS (from the same manufacturer) for a givensize can be produced over and over given the product demand. This has enabledrapid growth and adoption of SHS. As seen in Figure 2.6, between 2010 and 2016,the number of people gaining electricity access through SHS went from 7.2 to 23.5million people [17]. As of 2017, off-grid solar solutions have provided improvedenergy access to 73 million households [21].

2.5. PRESENT ISSUES WITH SHS-BASED ELECTRIFICATIONThe most directly significant barrier that prevents widespread adoption of SHS is theunit system cost. Technologically, this can be most directly addressed only in the form ofdeclining component costs. From a business point of view, it comes down to innovativebusiness models tailored to suit the target region and community. From a policy pointof view, subsidies could also play a role. These considerations are outside the scope ofthe work discussed in this dissertation. However, even if the adoption of present SHSwere to scale to a billion units radically, there are still two major drawbacks that will holdback existing SHS from being the perfect pathway towards universal electrification. Theseissues are discussed in this section.

2.5.1. CLIMBING THE ELECTRIFICATION LADDEROwing to its standalone nature, SHS is dimensioned to cater to a particular level of loaddemand. Dimensioning or sizing is often a critical exercise; an oversized SHS leads toan expensive system leading to wastage of energy and unaffordability to a considerableportion of the target population, while an undersized SHS leads to insufficient poweravailability that may cause user dissatisfaction and lower technology adoption rates.Therefore, SHS need to be optimally sized for a given load demand [30]. However, asis clearly evidenced in literature and observed in the field, one of the consequencesof providing access to electricity is the increase in electricity demand itself. That is,the demand for electricity often increases with time [31, 32, 27]. Therefore, in time(sometimes, in less than a year), the perfectly dimensioned SHS becomes obsolete anduntenable to cater to the growing demands of the user. In other words, a particular SHSsize is insufficient to enable the climb up the electrification ladder.

Figure 1.8 illustrated the concept of the electrification ladder. As a person or householdmoves up the electricity ladder, the quality and quantity of electricity demand increases,which can also be viewed as movement across the tiers in the context of the MTF. State-of-the-art SHS typically supply up to tier 2 level of electricity, and sometimes to tier 3.Even though a dedicated SHS for tier 3, 4 or 5 could be custom made, often it is not thebest solution because: (i) it would be at the cost of discarding the previously perfectly

2.5. PRESENT ISSUES WITH SHS-BASED ELECTRIFICATION

2

33

dimensioned lower tier SHS, as households seldom demand a direct tier 4 or 5 level accessand usually climb up the electrification ladder (ii) at higher levels of electricity demand,the storage costs would significantly drive up the total system costs [30, 33].

In general, the level of electrification is often a dynamic requirement as the electricityneeds increase with time. The current SHS-based electrification is largely inflexible tothis requirement. Appliances for productive use of energy and communal loads, whichare also usually commensurate with tiers 4 and 5 of electricity access, can also not bepowered with typical state-of-the-art SHS power levels.

2.5.2. THE PARADOX OF SHS-BASED ELECTRIFICATION - WATT ’S THE MAT-TER!

State-of-the-art SHS are usually limited to 100 Wp. This leads to a paradoxical predica-ment as shown in the conceptual illustration in Figure 2.9, which maps the appliancecosts versus power rating space. The highlighted region represents the range of the typicalSHS in terms of power delivery and the usual DC appliances used in such an SHS. Theseappliances are the so-called super-efficient DC appliances.

100 W

$200State-of-the-art SHS range

Cos

t

Power

Figure 2.9: A concept graph illustrating the mapping between power rating of the various home appliances andtheir costs. Note: the price level is just an indicative number, as the prices differ greatly across vendors andgeographies.

Horizon appliances like washing machines are definitely outside the scope of thepresent day SHS, both in terms of power and in terms of price, as denoted by the iconin the upper right corner in the graph. However, a 70 W laptop charger can be easilypowered by a 100 Wp SHS, while the appliance (laptop) itself might be well outside the

2

342. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

range of typical appliances afforded by an SHS consumer. Ironically, some applianceslike water kettles and rice cookers are already cheap enough (sub $5–10) for low-incomecommunities to buy, but cannot be powered by the SHS due to the high-power ratingof the appliances (e.g., a kW or more for the water kettle). In a Cambodian field studycarried out as part of this PhD project in 2016, inadequate power level has already beennoted as a shortcoming of the present SHS. The local users expressed interest in poweringwater kettles and rice cookers with their SHS. The field study was conducted to mainlyfind out the future needs of SHS users in rural Cambodia. Figure 2.10 shows photos fromthe fieldwork [27].

(a) Simulation game to engage the SHS users on their futureelectricity needs and appliance wishlist

(b) Visual representation to understand the ex-pected times of the future appliance usage

Figure 2.10: Photos from the fieldwork carried out in Cambodia as part of this PhD project [27]. Photo credit:Thomas den Heeten.

SILVER LINING OR RISK OF STAGNANCY?

Considering the above two issues, therefore, the biggest criticism for SHS comes in theform of its impermanence as an electrification pathway. It has been often regarded asmerely a short-term solution [34]. Despite the undeniable momentum in accelerating upto tier 2 level electricity access, SHS has been largely seen as a stopgap solution to providelighting and some basic appliances until the advent of the national grid.

The deployment of SHS leads to the unlocking of the latent electricity demand ofthe consumers, thereby paving the way for the need for higher power solutions likemicrogrids and grid extension. On the one hand, the silver lining is that an increaseddemand might justify the installation of higher power systems like centralized microgridsand grid extension. On the other hand, the delays and other roadblocks associated withgrid extension and microgrid installation might keep the consumers locked in at a certaintier of electricity access.

In general, from lighting a million to empowering a billion, solar home systems intheir present form can only take us part of the way.

2.6. THE HOLY GRAIL OF UNIVERSAL ELECTRIFICATION

2

35

2.6. THE HOLY GRAIL OF UNIVERSAL ELECTRIFICATIONNow that the three electrification pathways have been discussed in detail along with thechallenges and opportunities they face, we can think of the ideal pathway for achievinguniversal electrification. In line with the scope of this dissertation, the focus is on off-gridsolar-based electrification in this discussion.

Figure 2.11 shows the different electrification pathways mapped onto a plot of easeof electrification versus power level. Decentralized solutions like pico-solar and SHSperform much better on the ease of electrification, while the grid performs far better onthe power levels. In other words, current electrification pathways present an inherenttrade-off between the on-demand electrification versus on-demand power consumption.Centralized microgrids perform somewhat in the middle on both counts.

The idealpathway

central grid

centralized microgrid

SHS

pico-solar

Eas

eof

elec

trifi

catio

n

Power level

Figure 2.11: A concept graph illustrating the comparison between the different electrification pathways on aplot of the ease of electrification versus power demand.

The holy grail of universal electrification is represented in Figure 2.11 in the top rightcorner as the ideal electrification pathway that can provide rapid electrification whilecatering to on demand power consumption commensurate with higher tiers of electricityconsumption. This solution has to be as easy for rapid adoption as SHS while addressingthe current shortcomings plaguing SHS from becoming a truly sustainable electrificationsolution. The requirements from such an ideal off-grid solar solution that can overcomethe issues mentioned in Section 2.5 can be listed as follows.

1. Climbing up the electrification ladder. The solution should be expandable to

2

362. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

cater to growing household needs as well as scalable to accommodate increasingnumbers of the systems in the off-grid communities.

2. High power appliances. The solution should enable the use of high-power appli-ances.

3. Retrofitting. The solution should be able to work with other existing SHS.

4. Future-ready. The advent of the electricity grid in the future should preferably notrender the present solution completely obsolete.

2.6.1. SHS-BASED MICROGRID: A MEANS TO GET THERE?In the absence of fundamental technological breakthroughs that can radically affectcomponent costs or choice of energy technology, an SHS-based microgrid can be a meansto at least approach the ideal pathway. Figure 2.12 shows this concept, where a meshedDC grid can be created through the interconnection of existing SHS already operatingon DC. Such a microgrid can grow in time, as opposed to a centralized microgrid that isrelatively rigid with respect to initial system sizes.

Figure 2.12: Concept illustration of a bottom-up, scalable, interconnected SHS-based meshed DC microgrid[18]. The bottom house is shown without any SHS, indicating the possibility of purely consumptive housesjoining such a microgrid.

This concept has seen multiple proponents in recent literature [35, 36]. However,the existing projects in practice are mostly limited to tiers 1 to 2 with 12 V battery and12 V distribution. In order to enable higher tiers of consumption, such microgrids needto make use of higher voltage distribution as it minimizes distribution losses [37]. The

2.6. THE HOLY GRAIL OF UNIVERSAL ELECTRIFICATION

2

37

quantification of benefits from these microgrids has been covered in Chapter 6 and in arecent publication [18].

Such bottom-up, organically growing microgrids that are born out of the interconnec-tion of existing SHS can enable energy sharing. Additional advantages of such microgridswould include reduced system sizes as compared to standalone SHS and the capabilityto power appliances meant for productive use of energy. Furthermore, if and when thenational grid does arrive, this microgrid can already serve as a ready distribution grid(otherwise, grid extensions are merely reduced to providing electricity to just the periph-eries of villages, or the microgrids are left as stranded assets). Such bottom-up microgridsneed to be an essential piece of the electrification puzzle, which not only complementother electrification technologies and efforts but also become a logical transition stepfrom standalone systems like SHS. In this way, it also helps in preserving the sanctity ofthe SHS electrification pathway, which is already underway with remarkable momentum.

ADVANTAGES OF AN SHS-BASED, DECENTRALIZED, BOTTOM-UP MICROGRID

Compared to standalone electrification, following can be summarized as the main advan-tages of an interconnected SHS-based microgrid [18].

1. DC architecture. Exploiting the DC nature of its interconnected building blocks(DC), this can grow as a meshed DC microgrid.

2. Excess energy sharing. Interconnecting SHS enables excess energy sharing be-tween the households.

3. Reduced system size. A direct consequence of 2 is that the system sizes can be lowerthan the standalone case for meeting the same power availability requirement.

4. Productive use of energy. High power appliances enabling PUE can be easilysupported in an interconnected SHS microgrid. Productive use of energy cansupplement income-generating activities and therefore lead to a higher degree ofownership in the microgrid for the users.

5. Climbing up the electrification ladder. Climbing up the tiers of MTF will requiremuch lower increments in energy storage per household as opposed to a standaloneSHS climb up the tiers.

6. Retrofitting and ‘future-fitting’. Interconnected SHS-based microgrid not onlyhelps in reusing the existing SHS but also ensures that a DC distribution grid existsfor the national grid expansion if and when it reaches the target region.

Note: This is only one possible solution and by no means the silver bullet for attaininguniversal electrification. Nonetheless, given the current state of off-grid electrification, themassive investments in the off-grid sector and the inevitable climb up the electrificationladder, this seems to be the most promising route to scale electrification.

2

382. THE LONG ROAD TO UNIVERSAL ELECTRIFICATION: A CRITICAL LOOK AT PRESENT

PATHWAYS AND CHALLENGES

2.6.2. COMPARISON OF THE VARIOUS ELECTRIFICATION PATHWAYSTable 2.2 presents a qualitative comparison of the electrification pathways presented inthis chapter. The different aspects considered are in the context of universal electrificationonly. As discussed before and seen in Table 2.2, each of the three present-day electrifi-cation pathways have relative merits and demerits, while the decentralized SHS-basedmicrogrids generally perform well across the different aspects of comparison. The scala-bility and adoption aspect deserves a special mention, as that is a particular pain-pointin addressing the urgency of accelerated electrification. SHS can potentially continuewith their current momentum in adoption, but the level of electricity access will still belimited, as discussed in Section 2.5. Interconnected SHS-based microgrids, on the otherhand, can scale just as wide as SHS, but also grow in size due to interconnectivity.

Table 2.2: A qualitative comparison of the different electrification pathways on various aspects.

Aspect SHS Centralized mi-crogrids

Central grid Decentralized SHS-based DC microgrids

MTF tiers Tier 2–3 Tiers 1–5 (fixed) Tiers 1–5 Tiers 1–5, flexible

Implementationease

Very high Medium Low High

Scaling the electrifi-cation ladder

Very low Low High High

AC/DC DC AC, DC, hybrid AC DC

Communityloads/PUE

No Yes if planned Yes Yes, depends on themicrogrid size

Scalability & adop-tion

Very high Limited Limited High

DECENTRALIZED PARADIGM: HISTORY REPEATS, BUT TIME TO LEAPFROG INTO THE FU-TURE

The dawn of the electrical era, in fact, saw decentralized generation plants, together withbatteries supplying electricity via DC grids to nearby areas [38]. Then came along the ACgeneration and distribution system, with its technological advances at the time, and thecentralized electrification paradigm, which is still a mainstay of the electricity networksaround the world. However, when it comes to the 21st-century goals of electrification withits contemporary opportunities and unique challenges, we have distinct technologicaladvantages with decentralized DC electrification.

From power electronics to decentralized solar PV generation and battery storage tosuper-efficient DC appliances, the advances in technology can allow us to leapfrog to-wards a cleaner, sustainable electricity network based on DC when considering the daunt-ing task of universal electrification. Additionally, accelerating towards off-electrificationallows to leapfrog towards a bottom-up and decentralized electricity network paradigm.The time is opportune to make significant strides in electrification targets and ‘power’through towards SDG 7. Therefore, bottom-up, decentralized DC microgrid based oninterconnected SHS is not just an alternative to consider but also a logical choice.

2.7. CHALLENGES FOR THE SHS-BASED ELECTRIFICATION VISION

2

39

2.7. CHALLENGES FOR THE SHS-BASED ELECTRIFICATION VI-SION

The vision of SHS-based electrification described in Section 2.6.1 requires a few technicalchallenges to be solved first for SHS design at the lower tiers of electrification before think-ing of the integration of SHS into a microgrid network for higher tiers of electrification.This necessitates following challenges to be tackled, as also described in Section 1.2.

Knowledge of the load profiles Load profile is the basic first step needed when thinkingof optimally designing an off-grid system while reducing costs and maximizing poweravailability. Therefore, for an off-grid system like SHS, a better knowledge of the expectedload profiles can lead to better system design. Conversely, a poor knowledge of theexpected load profile can lead to oversizing of the system, leading to increased systemcosts, or undersizing of the system, leading to unacceptable power availability. However,load profiles are a scarce commodity for system designers in the context of electrification,especially when dealing with hitherto un(der-)electrified target regions.

Battery lifetime estimation Battery is a vital system component in an SHS owing to itsrelatively low lifetime with high costs amongst all SHS components. Along with intialinvestment or upfront costs, battery costs recur in the form of replacements during theSHS lifetime. As battery lifetime also depends on battery size, a reasonable estimate ofbattery lifetime at the SHS design stage can equip the system designer to balance theupfront and replacement costs based on the specific project application. However, batterylifetime needs to be estimated at the SHS design stage taking into account the expectedbattery cycling for a given level of electricity demand.

Optimal system sizing Given a particular load profile or electricity demand, an optimalsystem size will maximize the power supply availability, maximize the battery lifetime,while minimizing the total system upfront costs and excess power production. This turnsinto a multi-objective optimization problem. Therefore, the challenges related to knowl-edge of load profiles, battery lifetime estimation at the SHS design stage, and optimalSHS sizing would be tackled first in Chapters 3, 4, and 5, respectively. Consequently, thebenefits for going towards a decentralized DC microgrid based on SHS interconnectionare explored in Chapter 6. Optimal microgrid topologies are investigated in Chapter 7using a geographic information system (GIS)-based approach.

2.8. CONCLUSIONThe 3 different pathways for electrification, viz., grid extension, centralized microgrids,and standalone solar-based solutions like SHS, each come with their own set of limitations.SHS have been a promising solution so far in achieving up to tier 2 level electrification.However, in its current state, SHS can only light the way but is far from being the elec-trifying solution to achieve universal electrification. While decentralized solutions likepico-solar and SHS can provide on-demand electrification, the central grid can providehigher levels of power commensurate with tier 4 and tier 5. For climbing up the electrifica-

2

40 REFERENCES

tion ladder and enabling the use of high power appliances, an interconnected SHS-basedDC microgrid can overcome many of the limitations of SHS as well as centralized micro-grids. Such a bottom-up microgrid could well be the missing link between the presentelectrification pathways, forming an integral part of the universal electrification jigsaw.

REFERENCES[1] International Energy Agency, World Energy Outlook 2018. IEA, 2018.

[2] International Energy Agency, Energy Access Outlook 2017 - from poverty to prosperity.IEA, 2017.

[3] UN, “The Sustainable Development Goals Report,” tech. rep., United Nations, 2018.

[4] B. Tenenbaum, C. Greacen, T. Siyambalapitiya, and J. Knuckles, From the bottom up:how small power producers and mini-grids can deliver electrification and renewableenergy in Africa. World Bank Publications, 2014.

[5] M. Bhatia and N. Angelou, Beyond Connection, Energy Access Redefined. The WorldBank, 2015.

[6] IRENA, “Off-grid renewable energy solutions to expand electricity access: An op-portunity not to be missed,” International Renewable Energy Agency, Tech. Rep.,2019.

[7] H. Louie, Off-grid electrical systems in developing countries. Springer, 2018.

[8] Gabriel Davies, Matt Tilleard and Lucy Shaw, “Private mini-grid firms deserve achance to compete against slow utilities in africa.” https://www.greentechmedia.com/articles/read/a-faster-path-to-rural-electrification#gs.4p99x3. Accessed: 2017-05-19.

[9] M. Franz, N. Peterschmidt, M. Rohrer, and B. Kondev, “Mini-grid policy toolkit,”EUEI-PDF, ARE and REN21, Tech. Rep., 2014.

[10] IEG, World Bank Group Support to Electricity Access, FY2000–FY2014. World BankGroup, 2014.

[11] G. Manchanda, “Augmenting the diffusion of solar home systems for rural electrifi-cation — An Indian perspective,” Master’s thesis, Delft University of Technology, theNetherlands, 2017.

[12] T. Levin and V. M. Thomas, “Can developing countries leapfrog the centralizedelectrification paradigm?,” Energy for Sustainable Development, vol. 31, pp. 97–107,2016.

[13] IEA, Africa Energy Outlook - A focus on energy prospects in sub-Saharan Africa. Orga-nization for Economic Cooperation and Development, International Energy Agency,2014.

REFERENCES

2

41

[14] NREL, “Microgrids.” https://www.nrel.gov/grid/microgrids.html. Accessed:2018-10-13.

[15] S. Buluswar, Z. Friedman, P. Mehta, S. Mitra, and R. Sathre, 50 Breakthroughs -Critical scientific and technological advances needed for sustainable global develop-ment. LIGTT, Institute for Globally Transformative Technologies, Lawrence BerkeleyNational Lab, 2014.

[16] IBRD | The World Bank Group, Mini-Grids and Gender Equality : Inclusive Design,Better Development Outcomes. World Bank Group, ESMAP and Climate InvestmentFunds, 2017.

[17] IRENA, “Measurement and estimation of off-grid solar, hydro and biogas energy,”International Renewable Energy Agency, Tech. Rep., 2018.

[18] N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Quantifying the benefits of a solar home system-based dc microgrid forrural electrification,” Energies, vol. 12, no. 5, 2019.

[19] H. Olk and J. Mundt, “Photovoltaics for Productive Use Applications. A catalogueof DC-Appliances,” Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ)GmbH, 2016.

[20] N. Narayan, Z. Qin, J. Popovic-Gerber, J.-C. Diehl, P. Bauer, and M. Zeman, “Stochasticload profile construction for the multi-tier framework for household electricityaccess using off-grid dc appliances,” Energy Efficiency, Nov 2018.

[21] Lighting Global and Dalberg Advisors, “Off-grid solar market trends report,” tech.rep., Lighting Global, Dalberg Advisors, GOGLA, ESMAP, 2018.

[22] Global Off-Grid Lighting Association, “Global off-grid solar market report - semi-annual sales and impact data,” tech. rep., GOGLA, Lighting Global and Berenschot,2018.

[23] d.light, “S30 Solar LED Lantern.” https://www.dlight.com/product/s30/. Ac-cessed: 2017-08-12.

[24] d.light, “X850 Solar Home System.” https://www.dlight.com/product/x850/.Accessed: 2017-08-12.

[25] GOGLA, “Powering Opportunity — The Economic Impact of Off-Grid Solar,” GlobalOff-Grid Lighting Association (GOGLA) and Altai Consulting, 2018.

[26] Good Solar Initiative, Quality Charter: Technical and Service Quality Standards forAccredited Solar Suppliers. SNV Netherlands Development Organisation, 2016.

[27] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrifica-tion,” in 2017 International Conference on the Domestic Use of Energy (DUE), pp. 8–15,April 2017.

2

42 REFERENCES

[28] Lighting Global, Bloomberg New Energy Finance and GOGLA, “Off-grid solar markettrends report 2016,” Bloomberg New Energy Finance and Lighting Global in coopera-tion with the Global Off-Grid Lighting Association (GOGLA), 2016.

[29] I. Bisaga, N. Puzniak-Holford, A. Grealish, C. Baker-Brian, and P. Parikh, “Scalableoff-grid energy services enabled by iot: A case study of bboxx smart solar,” EnergyPolicy, vol. 109, pp. 199–207, 2017.

[30] N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Estimating battery lifetimes in solar home system design using a practi-cal modelling methodology,” Applied Energy, vol. 228, pp. 1629 – 1639, 2018.

[31] M. Gustavsson, “With time comes increased loads: An analysis of solar home systemuse in lundazi, zambia,” Renewable Energy, vol. 32, no. 5, pp. 796–813, 2007.

[32] F. Benavente, A. Lundblad, P. E. Campana, Y. Zhang, S. Cabrera, and G. Lindbergh,“Photovoltaic/battery system sizing for rural electrification in bolivia: Consideringthe suppressed demand effect,” Applied energy, vol. 235, pp. 519–528, 2019.

[33] N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Exploring the boundaries of solar home systems (shs) for off-grid elec-trification: Optimal shs sizing for the multi-tier framework for household electricityaccess,” Applied Energy, vol. 240, pp. 907 – 917, 2019.

[34] D. Chattopadhyay, M. Bazilian, and P. Lilienthal, “More power, less cost: transitioningup the solar energy ladder from home systems to mini-grids,” The Electricity Journal,vol. 28, no. 3, pp. 41–50, 2015.

[35] B. Soltowski, S. Strachan, O. Anaya-Lara, D. Frame, and M. Dolan, “Using smartpower management control to maximize energy utilization and reliability withina microgrid of interconnected solar home systems,” in 7th Annual IEEE GlobalHumanitarian Technologies Conference, GHTC 2017, pp. 1–7, 2017.

[36] S. Groh, D. Philipp, B. E. Lasch, and H. Kirchhoff, “Swarm electrification: Investigat-ing a paradigm shift through the building of microgrids bottom-up,” in DecentralizedSolutions for Developing Economies, pp. 3–22, Springer, 2015.

[37] M. Nasir, H. A. Khan, N. A. Zaffar, J. C. Vasquez, and J. M. Guerrero, “Scalable solardc micrigrids: On the path to revolutionizing the electrification architecture ofdeveloping communities,” IEEE Electrification Magazine, vol. 6, no. 4, pp. 63–72,2018.

[38] S. Mandelli, J. Barbieri, R. Mereu, and E. Colombo, “Off-grid systems for rural elec-trification in developing countries: Definitions, classification and a comprehensiveliterature review,” Renewable and Sustainable Energy Reviews, vol. 58, pp. 1621–1646,2016.

3LOAD PROFILE CONSTRUCTION

If knowledge can create problems, it is not through ignorance that we can solve them

Isaac Asimov

ABSTRACT

To improve access to electricity, decentralized, solar-based off-grid solutions like SolarHome Systems (SHS) and rural microgrids have recently seen a prolific growth, as dis-cussed in Chapter 2. However, electrical load profiles for these systems, usually the firststep in determining the electrical sizing of off-grid energy systems, are often non-existentor unreliable, especially when looking at the hitherto un(der)-electrified communities.This chapter aims to construct load profiles at the household level for each tier of elec-tricity access as set forth by the Multi-tier Framework (MTF) for measuring householdelectricity access. The loads comprise dedicated off-grid appliances, including the so-called super-efficient ones that are increasingly being used by SHS, reflecting the off-gridappliance market’s remarkable evolution in terms of efficiency and price. This studyculminated in devising a stochastic, bottom-up load profile construction methodologywith sample load profiles constructed for each tier of the MTF. The methodology exhibitsseveral advantages like scalability and adaptability for specific regions and communitiesbased on community-specific measured or desired electricity usage data. The resultingload profiles for different tiers shed significant light on the technical design directionsthat current and future off-grid systems must take to satisfy the growing energy demandsof the un(der)-electrified regions. Finally, a constructed load profile was also comparedwith a measured load profile from an SHS active in the field in Rwanda, demonstratingthe usability of the methodology.

This chapter is based on: N. Narayan, Z. Qin, J. Popovic-Gerber, J.C. Diehl, P. Bauer, & M. Zeman. (2018).Stochastic load profile construction for the multi-tier framework for household electricity access using off-gridDC appliances. Energy Efficiency, 1-19.

43

3

44 3. LOAD PROFILE CONSTRUCTION

OUTLINE

This chapter contains 5 sections. Section 3.1 introduces the requirement of load profileswithin off-grid systems and the development in super-efficient DC appliances for off-gridelectrification. Sections 3.2 discusses the literature review and the relevant load profileparameters. Section 3.3 describes the methodology used to construct the load profiles.Section 3.4 discusses the results and also presents a comparison of the constructed tier 1load profile with SHS load data obtained from the field in Rwanda. Finally, Section 3.5concludes the chapter with additional recommendations related to load profile construc-tion.

3.1. INTRODUCTION

3.1.1. MULTI-TIER FRAMEWORK FOR HOUSEHOLD ELECTRICITY ACCESS

Section 2.1.1 already introduced the concept of the MTF. The central idea behind the MTFis to view electrification in terms of the quality and quantity of the energy services suppliedto the user. The underlying premise is that different energy service attributes, as listedin Table 2.1, define the electricity usage that satisfies the various human needs. Humanneeds broadly encompass the following categories: (a) lighting, (b) entertainment andcommunication, (c) food preservation, (d) mechanical loads and labor-saving, (e) cookingand water heating, (f) space cooling and heating [1]. The household level energy useis bound to be increasing with time as more households rise out of (energy) poverty[2]. Moving up the tiers makes more and better quality energy services available thatcould satisfy more of these human needs. The notion of this tier-based framework isindependent of the type of technology that enables the electrification. However, asdiscussed in Chapters 1 and 2, the focus of the energy technology is confined to SHS inthis dissertation.

3.1.2. OFF-GRID APPLIANCES

Since the dawn of electricity, the electricity appliance market has always been under aconstant state of evolution. However, until recently, most major advents of electricalappliances and devices in the developing world have largely been on the heels of theirsuccess in the developed world. Examples include mobile phones and LED lights. Thistrend has now been broken, and a number of dedicated, so-called super-efficient off-gridappliances are being designed specifically for the off-grid markets [4, 5]. Examples includeelectric fans, TVs [6], and refrigerators, all of which can work easily with off-grid PV onDC as well as consume just a fraction of the energy costs of the traditional, mainstreamcounterparts. Moreover, high quality off-grid products tailor-made for the low-incomemarkets combined with innovative business models and falling component costs haveenabled the rapid market growth.

Envisioning universal electricity access by 2030 would require the support of off-grid,standalone systems as well. The use of super-efficient appliances will greatly help in thisregard, as seen in Figure 3.1. The higher cost of bundling super-efficient appliances ismore than compensated by the lower cost of the PV and battery due to reduced systemsize needed to deliver the same energy services. In this particular example, an averageannual cost reduction of 32% can be achieved [3]. In general, coupling super-efficient

3.1. INTRODUCTION

3

45

I II

100

200

300

400

500

600

$/ye

ar/h

ouse

hold

(201

6)

Increase in

appliance $Decrease in

system $

Appliances

SHS

Figure 3.1: Impact of using super-efficient appliances in overall system costs. Assuming universal electricityaccess by 2030, annual average cost per household powered by SHS and using four light bulbs, a television, afan, a mobile phone charger and a refrigerator. I: SHS with standard appliances; II: SHS with super-efficientappliances. Use of super-efficient appliances leads to an overall decrease in total costs. Adapted from [3].

devices has been seen as a means to delivering the same, if not better energy services atlower costs, while also increasing the momentum of the energy access efforts [7].

Moreover, many of these appliances can be used for their productive use of energy,i.e., using the appliances for improving the productivity and supplementing income[8]. Productive use of energy (PUE) is different from the usual, consumptive use ofenergy, in that the productive use is labor-saving and supplements the income generationcapabilities of the user. Some examples are in small enterprises (sewing machines, powerdrills) and farming (solar pumps). The biggest advantage of PUE comes in the form ofbenefits towards both poverty reduction as well as stability and viability of energy supply,due to increased ability to pay for the energy services, and less reliance on subsidies [9].However, not all dedicated off-grid appliances available today can be classified in thesuper-efficient category. For instance, washing machines and air coolers still have a longway to go in terms of efficiency improvements.

It is interesting to note that super-efficient appliances are also being promoted fromthe policy side by governments, and are not necessarily limited to the off-grid sector.For example, by providing production subsidies to appliance manufacturers, the SuperEfficient Equipment Program (SEEP) aims at reducing energy consumption in Indianhouseholds [10]. International collaborative programs like CLASP are also working to-wards improving energy efficient appliances.

3.1.3. IMPORTANCE OF LOAD PROFILES

A load profile can be defined as the power demand of an energy system mapped over time.A load profile not only captures the energy demand of the user but also serves as a vitalinput to the electrical system design. Especially in the case of off-grid power systems like

3

46 3. LOAD PROFILE CONSTRUCTION

the SHS, reliable apriori knowledge of the load profile is extremely helpful in the electricalsizing the system (i.e., deciding the PV rating and the battery capacity).

In fact, load profile, even if coarsely estimated, is almost always the starting point in anoff-grid, standalone PV system design [11]. Load profiles can have a profound impact onthe performance as well as design decisions in off-grid systems [12]. A better knowledgeof load profile makes for a more optimal off-grid electrical system design. Conversely, thelack of an appropriate load profile leads to either oversizing or undersizing the system,thereby causing an unhealthy trade-off between system costs and power availability [13].

An over-utilization will result in frequently empty batteries and loss-of-load events.Particularly in the context of first-time SHS users, this may result in loss of faith andtrust in solar-based electrification, as was experienced during the fieldwork done in ruralCambodia related to the work reported in [13]. On the other hand, under-utilizationwould have a detrimental financial impact. This is because the user would pay more foreffectively utilizing only a fraction of the power generation potential of the system. Interms of the levelized cost of electricity (LCOE), a grossly underutilized system wouldresult in a high LCOE. In a market segment where purchase power, cost price, and profitmargins are sensitive parameters, a high LCOE is definitely unattractive, whether or notthere are subsidies in play.

3.1.4. NEED FOR LOAD PROFILE CONSTRUCTIONThe following points necessitate the requirement for load profile construction.

• Difficult to estimate It is tough to estimate the energy consumption behaviour inoff-grid communities if the electricity access has hitherto been limited.

• Starting point The load profile is the starting point in off-grid system design. In theabsence of existing load profiles, as is the case with most off-grid locations, reliableload profile construction is crucial.

• Growth enabling Load profile construction for not just the present but an estimatedfuture usage would benefit the off-grid system designers to enable the growth ofthe energy consumption in their off-grid systems.

3.1.5. HIGHLIGHTSFor the goal of achieving an optimal SHS design, the work described in this chapterendeavours to construct load profiles by mapping the energy usage for various tiers ofelectricity access. While multiple studies in the past have discussed load profile construc-tion (as discussed in Section 3.2.1), this chapter focuses specifically on stochastic loadprofile construction for the various tiers of the MTF for electricity access. Following arethe main highlights of this chapter.

1. A fully adaptable, scalable, and bottom-up load profile construction methodologyis presented, which can be customized for different regions and user groups.

2. The constructed load profiles include the latest trends in dedicated off-grid appli-ances for each tier of the MTF for household electricity access.

3.2. BACKGROUND

3

47

3. Off-grid appliances for productive use of energy are included while constructingthe load profiles.

4. The stochastic load profile construction model is validated through a comparisonwith a measured one from an active SHS in the field.

5. The impact of load profiles on off-grid system design is discussed.

3.2. BACKGROUNDThis section discusses a brief literature review on the existing load profile constructionmethodologies, and details the load profile parameters and the types of off-grid electricalappliances considered in this study.

3.2.1. LITERATURE REVIEWLoad profile construction can be a complex task for the rural off-grid scenario. Theelectrical consumption of the target users is often increasing with time, as the process ofelectrification itself contributes to greater energy needs because of improvement in livingstandards [14]. It is therefore important for rural off-grid initiatives to enable communitiesto move beyond basic lighting and phone charging [15].

However, the complex causality of electricity access with socio-economic develop-ment significantly impacts energy modelling approaches, while most electrification im-pact assessments estimate a linear growth in electricity demand [16]. Limited knowledgeof electricity demand can negatively impact off-grid system dimensioning. Similarly, ithas been noted that most studies do not assume a dynamic demand over the years, anddo not consider the evolution of the future load demand, which severely undermineslong-term planning [17]. As seen in [17], 74.4% of energy demand forecasting approachesadopted in the reported case studies consider no evolution in the energy demand.

Fortunately, thanks to its multi-layered approach, the MTF for household electric-ity access can be instrumental in overcoming some of these problems. The tier-basedelectrification makes it easier to categorically cater to different tiers of energy usages.Additionally, while the jump from, say, tier 1 to tier 2 might happen quickly for certainhouseholds, it would still take much longer to reach tier 4 or 5. Using MTF as a categori-cal sieve for modelling energy demand and consequently designing electrical systemsstill makes a better case for the future than assuming a fixed demand. Moreover, use ofinterviews and surveys can be used additionally to complement and calibrate the loadprofiles based on the MTF, thereby making them more robust as well as customized tolocal contexts.

In the case of rural (off-grid) electrification projects, use of only interviews to modelthe energy demand may not be the best way forward. Currently, energy demand estima-tion is often carried out solely through interviews and surveys due to lack of historical data.However, research-surveys for predicting electricity consumptions can be error-prone. Ashigh as 426 Wh/day per consumer of mean absolute error has been observed over a studyof eight rural mini-grids in Kenya [18]. In [19], the load profiles are compared in a mini-grid from interviews as well as measured data, concluding that purely interview-baseddata falls short in accurately estimating energy needs. Therefore, interviews, when used

3

48 3. LOAD PROFILE CONSTRUCTION

as the only tool, need not be the best indicators of actual load consumption, especially ifdelicate system design and sizing is going to be based on the load profiles constructedfrom interview data. However, a stochastic load profile construction approach can gaingreatly from interview data in terms of calibration and limits.

Existing energy modelling methods have in general been investigated by [20] in thecontext of developing economies. Econometric (macro-scale) and end-use are the twomost common approaches being used, with the latter able to produce more realisticprojections. Given the task of estimating household energy demand and constructingload profiles for the same, modelling end-use consumption is more suited. In [21], twomain approaches in modelling user data and generating load profiles are identified, viz.,top-down and bottom-up approaches.

Top-down approaches work on aggregate data, which does not distinguish energy con-sumption due to individual users. Pattern recognition and clustering methods to generateload profiles as well as for short/mid-term load forecasting are used in [22]. However, thisis oblivious to appliance level data. Similarly, clustering-based methodologies are used toperuse large datasets and construct ‘new’ load profiles based on the existing data in [23]and [24]. Load aggregation and sampling times of measured data to determine optimalsampling times and aggregation levels are discussed in [25].

Based on the top-down approach, these models lack appliance level data and are notspecifically suitable for off-grid applications. Reliance on historical aggregate data is adrawback of these models, as they are intrinsically incapable of modelling technologicaladvances that are discontinuous in nature [21]. Therefore, top-down approach is deemedinadequate to be used for modelling the energy consumption in the form of load profilesfor the purpose of electrification of remote, off-grid households that are hitherto un(der)-electrified, while also taking into account the latest developments in dedicated off-gridappliances.

On the other hand, bottom-up approaches work with individual users, and often evenwith appliance level data. Bottom-up approaches can be classified as Statistical Methods(SMs) or Engineering Methods (EMs). While SMs use regression analysis or artificialneural networks to estimate end-user energy consumptions, and use historical data, EMsaccount for energy consumption at individual user levels on the basis of power ratings anduse of equipment [21]. These do not necessarily rely on historical or measured data, butmeasured data can be used for calibration. To construct load profiles while incorporatingthe off-grid appliances, only the EM methods can be used. Moreover, as one goes to asmaller scale, like a household level, load profiles need to be more appliance oriented thanwhat the top-down approaches can offer. This is because demand and supply needs to beoptimally matched in planning and operation. The maximum demand and variabilitydecreases with increasing number of appliances and users [26]. Therefore, for smallermini-grids and SHS, the peaks are even more disruptive and the demand and supplymatching even more difficult. Consequently, a good peek into the load consumptionprofile at the design stage is crucial. Therefore, a bottom-up approach is chosen forstochastic load profile construction in this study.

Bottom-up approach is used in [27], which uses datasets for hundreds of Finnishurban households to construct load profile from each load behaviour. Underlying loadsand appliances were obtained from other studies at the time (2006). Probability factors

3.2. BACKGROUND

3

49

were taken from public reports and other available data. However, this study was primarilyin the urban scenario, with contemporary loads, and considering hourly mean powerlevels.

In [28], a bottom-up approach to construct load profiles specifically for off-grid areasis discussed. While [28] comes closest to this study for creating a bottom-up methodologyfor constructing load profiles for rural off-grid systems, there are additional aspects thatare considered in the study described in this chapter, which will be incorporated in themethodology described in Section 3.3. These are described below.

(a) Special focus on dedicated, super-efficient, off-grid (DC) appliances and relatedtrends.

(b) Formulating an analytical approach to include the coincidence factor as a designparameter for stochastic load profile construction.

(c) Each load per tier constrained by a maximum usage per day.

(d) Use of a measured load profile from an SHS in Rwanda to compare and validate theconstructed load profile.

(e) Constructing daily load profiles for an entire year so as to also account for inter-dayvariations. The methodology could be further complemented with interviews anddata monitoring for seasonal recalibration if necessary.

3.2.2. LOAD PROFILE PARAMETERSThe important load profile parameters that referred in this work are discussed in thissection. Additionally, implications of the load parameters on the off-grid energy systemare also discussed.

PEAK AND AVERAGE LOADS

Peak load is the maximum power value the load profile takes over the period of considera-tion, while the average load is the mean load power. The electrical dimensioning of theenergy system must be able to support the peak load power. In SHS, this is usually donewith adequately rated power electronic converters. Information on average loads is initself insufficient to appropriately size the systems and often error-prone [29].

ENERGY DEMAND

The energy demand is the integral of instantaneous power demand, i.e., the summationof all load consumption events during the period of consideration. This has implicationson the amount of electrical energy the generator like PV module has to produce in thesystem.

E =T∫

0

PL(t )d t (3.1)

Where E is the total energy demand, PL(t ) is the instantaneous load power, and T is thetime interval under consideration. Consequently, the mean daily energy, which is definedas the average daily energy consumption over the given year, can be obtained as follows.

3

50 3. LOAD PROFILE CONSTRUCTION

Edaily =

∑365i=1

24h∫0

PLi (t )d t

365(3.2)

Where Edaily is the mean daily energy, PLi (t ) is the instantaneous load power on day i .

LOAD FACTOR

Load factor is defined as the ratio of the average load power to the peak load power asshown in Equation 3.3. Higher the load factor, flatter the load profile, indicative of a loaddemand with fewer variations. On the other hand, a low load factor is indicative of highvariability in the load profile.

Load Factor = PL,Avg

PL,Peak(3.3)

Where PL,Avg and PL,Peak are the average and the peak power values of the load profileover a particular day.

COINCIDENCE FACTOR

Coincidence factor (CF) is defined as the ratio of the peak load power in the load profileto the total rated power of all the installed appliances in the energy system, as shown inEquation 3.4 [19]. It is a measure of the likelihood of all the loads constituting the loadprofile functioning simultaneously. This can hugely impact the electrical dimensioningof the system.

CF = PL,Peak

PL,Tot(3.4)

Where PL,Tot is the total installed load.

3.2.3. TYPES OF APPLIANCESSection 3.1.2 has already introduced the so-called super-efficient appliances increasinglybeing produced and used in the rural, off-grid regions. [30] reports a dedicated surveyon off-grid appliances to investigate the appliance-products reflecting the greatest off-grid consumer demand, as well as driving the greatest electricity access. This survey,carried out in Africa, Latin America, and South Asia, considered a total of 19 appliancesfor off-grid household usage. The appliances considered in the survey have been listed inTable 3.1.

The table also mentions the typical power ratings and the needs that these appliancescan fulfill. This survey and its results reported in [30] have been taken as the base for theselection of the loads that can meet the energy needs and can help in the construction ofthe load profiles. These loads can now be used in constructing the load profiles for thevarious tiers of electricity access as outlined by the MTF.

3.3. METHODOLOGYThis section details the overall methodology used for the stochastic construction of theload profiles, along with the mathematical model that forms the core of the methodology.

3.3. METHODOLOGY

3

51

Table 3.1: List of off-grid appliances as ranked in [30]. The ‘Needs’ column refers to the same human needs a to fas mentioned in Section 3.1.1.

S. No. Appliance Needs Typical Rating [W]

1 LED Lighting a 1 - 52 Mobile charging/ banks b 3 - 203 Television b 10 - 504 Radio b 2 - 55 Fridges c 40 - 4006 Fan f 15 - 1007 Laptop b 30 - 1008 Solar water pumps d 40 - 8009 Tablets b 12 - 50

10 Rice cooker e 200 - 25011 Clothes iron d 150 - 200012 Grinders d 750 - 100013 Hand power tools d 100 - 100014 Hair clippers d 15 - 5015 Small speaker systems b 5 - 1016 Rice mills d 200 - 50017 Sewing machines d 40 - 10018 Soldering iron d 20 - 6019 Tea kettles e 100 - 800

3.3.1. LOAD CLASSIFICATION

Out of the 19 listed off-grid appliances listed in Table 3.1, a total of 16 appliances wereconsidered in this study, with an air cooler as an additional appliance. These loads arethen classified into the 5 tiers (T-1 to T-5) of household electricity access as outlined bythe MTF (Table 2.1). Every tier is a superset of the preceding tier with more appliances.

This classification is captured in Table 3.2, where the considered load appliancesare listed along with their power rating, quantity, and other operating constraints thatare used as inputs to the mathematical model, as described in Section 3.3.2. Table 3.2assumes typical values of operational constraints to demonstrate the applicability of thedescribed methodology. These values are expected to be customized per target region orcommunity.

Additionally, four dedicated appliances that contribute to productive use of energy(PUE) are considered as well in tier 5, viz., hand power tools, water pump, grinder/miller,and sewing machine. Note that the so-called super-efficient off-grid appliances areconsidered wherever applicable. However, some high power appliances, although DCand off-grid, are still in the process of undergoing advancements in the off-grid market,and therefore need not necessarily be ‘super-efficient’ like the LEDs and TV. For example,washing machine and air coolers. The appliances, and their specific datasheets, whereverapplicable, are mentioned in the Appendix at the end of this chapter.

3

52 3. LOAD PROFILE CONSTRUCTION

Table 3.2: Various operational constraints and inputs to the load profile construction methodology. T-1:T-5 =Tier 1:Tier 5.

Load j Rating P [W] Tmin Tmax Tm [h] nmin nmax W1start

W1length

W2start

W2length

Quantity q [-]

T-1

T-2

T-3

T-4

T-5

[min] [min]T-1

T-2

T-3

T-4

T-5

[-] [-] time [h] time [h] T-1

T-2

T-3

T-4

T-5

LEDLight-ing

2 2 2 2 2 30 240 6 8 8 12 12 1 12 4 2 18 6 3 5 5 8 12

MobilePhoneCharg-ing

3 3 3 3 3 5 120 6 8 8 8 8 1 12 0 1 6 18 2 3 3 5 5

Radio 0 3 3 3 3 5 240 0 8 8 12 12 1 10 7 13 0 0 0 2 2 2 2Fan 0 15 20 35 35 5 600 0 8 8 12 16 1 10 7 12 0 0 0 1 2 4 4TV 0 12 18 29 29 5 240 0 8 8 12 12 1 10 7 7 17 6 0 1 1 2 2Fridge 0 0 54 54 54 5 30 0 0 3 8 24 5 15 5 19 0 0 0 0 1 1 1Tablet 0 0 18 18 18 5 120 0 0 6 12 12 1 10 0 1 6 18 0 0 1 1 1Kettle 0 0 0 400 400 5 15 0 0 0 1 1 1 8 7 14 0 0 0 0 0 1 1Laptop 0 0 0 60 60 5 240 0 0 0 6 6 1 10 0 1 6 18 0 0 0 1 1RiceCooker

0 0 0 200 200 30 30 0 0 0 1.5 1.5 1 3 10 2 17 3 0 0 0 1 1

ClothesIron

0 0 0 150 150 5 20 0 0 0 2 2 1 2 6 16 0 0 0 0 0 1 1

Washingma-chine

0 0 0 70 70 15 120 0 0 0 2 2 0 1 6 14 0 0 0 0 0 1 1

Aircooler

0 0 0 500 500 30 120 0 0 0 4 12 0 8 9 9 0 0 0 0 0 1 1

Powertools

0 0 0 0 100 5 60 0 0 0 0 10 2 10 6 14 0 0 0 0 0 0 1

Grinders/Millers

0 0 0 0 750 10 120 0 0 0 0 10 2 10 6 14 0 0 0 0 0 0 1

SewingMa-chine

0 0 0 0 40 5 120 0 0 0 0 10 3 20 6 14 0 0 0 0 0 0 1

WaterPump

0 0 0 0 750 5 30 0 0 0 0 4 0 2 5 12 0 0 0 0 0 0 1

3.3.2. MODEL PARAMETERSThe parameters of the mathematical model used in the methodology are defined inTable 3.3. The model assumes that a load can be operated in an allowed usage windowmultiple times for finite durations, as defined by the parameters and their constraints.The values for these parameters (including boundary values where applicable) along withdetails of the load appliance used for constructing the load profile can be seen in Table 3.2.It must be noted that these input values have a bearing on the model output, and canbe tailored by the system designer using the described methodology for the specific usergroup being catered to.

As the data resolution in this study is 1 minute, functioning time f (Table 3.3) is a1440 (1 day long) length vector in MATLAB having ones and zeroes (1s and 0s) to indicateactivity or inactivity of any load.

USAGE WINDOW

The idea of the usage window is illustrated in Figure 3.2. The concept illustration shows ausage window (W ) spanning 9 hours from 09.00 to 18.00. Two occurrences of variableduration of a load rated 150 W can be seen, from 11.00 to 12.00 and 15.00 to 18.00.Additionally, the cycle times (Ti ), number of instances (n), starting times (ti ) and power

3.3. METHODOLOGY

3

53

Table 3.3: Parameters used for the stochastic load profile construction.

Parameter Definition Notation Type

Load type Type of load in use. For example, TV, LEDlights, etc, chosen from the LCM

j Input

Total load type Total load types per category N InputRated power Rated power per load PCycle time Duration for which a load is operational T GeneratedMaximum Usage Maximum number of hours a load is al-

lowed to operate during the dayTm Input

Instances Number of times a load is operated in theallowed usage window

n Generated

Usage window The allowed time window within whichthe loads are expected to be used

W Input

Quantity of loads The number of loads of each type q InputStart time Start time of a load instance within the

usage windowt Generated

Load occurrence The i th occurrence of a load cycle i GeneratedDynamic Window Dynamic window calculated as more load

occurrences reduce the usage windowWdyn Generated

Functioning time Day long time interval showing the pre-cise time stamps when the load is activeor inactive

f Output

Peak window The usage window where multiple appli-ances can potentially function simultane-ously

Wpeak Generated

Coincidence Factor a design input between 0.2 and 1 that de-notes the likelihood of appliances func-tioning simultaneously

C F Input

rating (P ) have been marked.

PEAK WINDOW AND COINCIDENCE FACTOR

Peak window is that subset of the total allowed usage window that intersects with theusage window of other loads. This is an important input parameter for the model, asdepending on the coincidence factor, multiple loads will operate in the peak window atthe same time or around each other.

Figure 3.3 illustrates the concept of the peak window. The load usage window W j isshown for 3 different appliances W1 to W3, and the intersecting time duration Wpeak isshown. For tiers 4 and 5, lower powered applications like LED lighting, mobile phonecharging, and radio are not considered for determining the peak window.

Using coincidence factor The concept of coincidence factor (CF) described in Sec-tion 3.2.2 is an important load profile parameter, which can have implications on the

3

54 3. LOAD PROFILE CONSTRUCTION

0 3 6 9 12 15 18 21 240

50

100

150

200

Time [h]

Pow

er[W

]

W

n = 2

T1 T2

P

t1

t2

Figure 3.2: Concept illustration of load usage window (W ) and load occurrence during the day.

0 3 6 9 12 15 18 21 240

50

100

150

200

Time [h]

Pow

er[W

]

W2

W1

W3

Wpeak

Figure 3.3: Concept illustration of peak window (shaded region) as an intersection of 3 different load usagewindows.

off-grid system design to cater to the load profile. The coincidence factor measure is usedas an attribute that impacts the probability of a load occurrence in the peak window. Theprobability of load occurrence times ti j within the peak window is considered to followa normal distribution with the mean centered around the middle of the peak window.Equation 3.5 describes the normal probability density function with a mean µ and stan-dard deviation σ. It is known that 99.7% of the values drawn from the normal distributionare within 3σ distance from the mean.

P (x) = 1

σp

2πe−(x−µ)2

/2σ2

(3.5)

Figure 3.4 illustrates the normal probability distribution of load occurrence beingsuperimposed on the peak window. A high coincidence factor would imply a much nar-rower distribution compared to a lower coincidence factor. Minimum coincidence factor

3.3. METHODOLOGY

3

55

µ−3σ 3σ−3σ 3σ

Wpeak

CFmin

CF = 1CFmin < CF < 1

Figure 3.4: Normal probability distribution of load occurrence in the peak window.

gives rise to a normal distribution that just fits within the window, with the peak windowboundary marking the 3σ limits for the normal distribution. Maximum coincidence factorof 1 would imply a distribution with σ= 0 and therefore the occurrence guaranteed at thecenter of the peak window.

Additionally, this is used to derive a relationship between CF and sigma as shown inEquation 3.6. C Fmin is fixed at 0.2 in this study.

σ= 1−C F

1−C Fmin× Wpeak

6(3.6)

RANDOMNESS AND CONSTRAINTS

As the model stochastically builds a load profile in a bottom-up manner by constructingload usages during the day, randomness was incorporated in the model. There are 3 mainparameters that are quantified randomly withing the usage windows. These are: cycletimes, starting times within the usage window(s), and the number of instances of loadoccurrence.

Moreover, the starting time for the specific occurrence in the peak window is handledusing a normal distribution and the coincidence factor as explained in Section 3.3.2. Theload profile construction efforts described in this chapter have been executed in MATLAB,and the randomness has been achieved with the help of the internal random numbergenerator of MATLAB.

Additionally, there are several constraints that the various parameters are bound by.These are listed as follows.

1. i ∈ [1,n j ]

2. j ∈ [1, N ]

3. n j ∈ [n jmin ,n jmax ], where n jmin and n jmax are the lower and upper bounds of thenumber of instances of load occurrence for load j in a window respectively.

4. Ti j ∈ [T jmin ,T jmax ] where T jmin and T jmax are the lower and upper bounds of thecycle times respectively.

5.∑n j

i=1 Ti j ≤ Tm, j

3

56 3. LOAD PROFILE CONSTRUCTION

6. ti j ∈W j ∀ i j . Additionally, ti j +Ti j ∈W j

These constraints change across the different load categories. The values of each ofthese constraints for the various loads and different tiers have been captured in Table 3.2.

j

n j ,min

n j ,max

W j

T j ,min

T j ,max

q j

P j

Tm, j

C F

Randomn j

RandomTi j

DetermineWpeak

Randomti j

ti j +Ti j

UpdateWdyn

UpdateTtotal, j

Tm, j >Ttotal, j

fits inWdyn

TruncateTi j

TruncateTi j

Update f j

L =∑Nj=1 f j P j

Executed∀q, j ?

DetermineWpeak no

yes

no

yes

Repeat for remaining i , q

yes

no

Inputs Generated

Phase I

Generated

Phase II

Generated

Phase III Phase IV

Output

Figure 3.5: Flowchart illustrating the stochastic load profile estimation for one day. The process is repeated forthe whole year across every tier.

3.3. METHODOLOGY

3

57

3.3.3. STOCHASTIC LOAD PROFILE MODELThe stochastic load profile construction methodology is illustrated through the flowchartpresented in Figure 3.5. The methodology is executed in MATLAB in four distinct phases.They are as described below.

1. Phase I: Stochastic generation In this phase, for every load j , the number of in-stances n j of load occurrence during the day is randomly generated using therandom number generator in MATLAB. Then, for each load instance occurrence i , acycle time Ti j is randomly generated. The peak window Wpeak is also determined inthis phase. The various constraints, quantity of loads, rated power, and coincidencefactor (C F ) are also taken as inputs in this phase.

2. Phase II: Occurrence Distribution The occurrence time instance t of every loadinstance i is determined randomly in this step. Note that one occurrence is delib-erately considered inside the peak window, using normal probability distributionwith parameters derived from the coincidence factor, as explained in Equation 3.4in Section 3.3.2. The occurrence ti occurs inside the allowed window(s) W , sothat in general ti j ∈W j . The cycle time Ti is added to ti . At this time, Phase III isexecuted to validate the cycle time of the load. If valid, then Phase II is executedagain for the next load instance i . If invalid, then the cycle time is constrainedaccording to either of 2 criteria as mentioned in Phase III, and then Phase II isexecuted for the next load instance i . If i was already the same as n, then PhaseII is implemented for the next load of the same type if the quantity q > 1. If thefunctioning window f has been generated for all q then the execution proceedsto the next load j . For every execution of Phase II, the allowed window is shrunkbased on the cycle time occurrence of the executed load instance. Therefore anupdated, shrunken, dynamic window Wdynis available for the next load occurrence.The total usage hours of any load is also kept track of in this phase.

3. Phase III: Validation and Correction Phase III concerns with validation based on 2criteria, viz.,

∑n j

i=1 Ti j ≤ Tm, j and ti j +Ti j ∈W j . That is, the cycle time is truncatedif it spills out of the dynamic window. Alternately, it is truncated if the sum total ofall load occurrences is greater than the allowed usage hours, a constraint comingfrom the MTF. After every execution of Phase II, Phase III validates the result, andPhase II is executed again accordingly. The methodology thus operates betweenPhase II and Phase III until the functioning time f j is populated for each q of all j .

4. Phase IV: Aggregation At this point, the functioning windows f j have been stochas-tically generated for all the loads. Finally, the load profile is aggregated taking thequantity q of the loads and the power rating P into account as shown in Equa-tion 3.7.

L =N∑

j=1f j P j (3.7)

Where L is the final load profile vector of length 1440, owing to the 1-minute resolution off . The load profile generation is repeated 365 times for a yearly load profile.

3

58 3. LOAD PROFILE CONSTRUCTION

The procedure described is valid for the generation of the load profile for N loadswithin the user group. The same procedure can be followed with different inputs andconstraints to obtain the load profiles from various tiers. The reader is referred to Table 3.2to peruse the entire list of inputs and constraints as used in this study.

ASSUMPTIONS IN THE METHODOLOGY

There are a few careful assumptions in the presented methodology. These are:

1. No occurrence of the load takes place outside the usage window. Therefore, thechoice of usage window largely dictates the particular load’s contribution in theoverall load profile. The usage windows shown in Table 3.2 have been consid-ered as the typically expected values, which could be made more precise as usagepreference data is available from the field.

2. For some of the loads, like laptop, tablet, and mobile phone charging, the loads inthese cases specifically denote just the charging of the devices’ battery and not theusage of the loads itself.

3. The load power is constant throughout its occurrence and is equal to the ratedpower. The only exception to this is the refrigerator, as explained below.

It must also be noted that the methodology presented here is strictly for load profileconstruction, and not for load forecasting.

THE CASE OF THE REFRIGERATOR

Unlike the rest of the loads, the fridge’s energy consumption is rated differently andtherefore deserves special treatment. The power consumption behaviour of the fridge isnot taken to be at rated power at all times. This is because manufacturers often specify twoparameters, the rated power and the daily energy consumption under testing conditions.

For example, an off-grid DC-powered fridge is rated at 40 W with a daily energyconsumption of 114 Wh at 32.2C ambient temperature [31]. The daily energy mentionedtranslates to a steady consumption of 4.75 W, much lower than the rated power of 40 W.This is because the daily energy only denotes the consumption when the fridge is largelyin standby, and otherwise switches on due to self-checks. This neglects the switching onof the fridge due to external events like the door opening and new food addition, etc.

The methodology described in this study assumes the rated standby and self-checkconsumption without external events. Additionally, the load instances i occurring throughthe day are considered to be the external events causing the fridge to switch on andconsume 40 W for the cycle times Ti . Just as the other loads, the number of eventinstances will be constrained as i ∈ [1,n] and n ∈ [nmin,nmax].

3.3.4. ADVANTAGES OF THE METHODOLOGYThe load profile construction methodology developed in this study has several advantages,as listed below.

1. Scalability The load profiles constructed in this study were at a household level,given the background of electricity access and the MTF. However, as the method-ology is bottom-up, the same method can be used to scale up the scope from

3.4. RESULTS AND DISCUSSIONS

3

59

household to neighbourhood or even village level. The coincidence factor will needto be appropriately changed.

2. Randomness The stochastic nature of the load profiles embodies uncertainty, re-flecting human behaviour. For example, there is consistency with the usage win-dows for each load. However, the number of times of load usage and the cycletimes are still inconsistent within constraints. This allows for multiple load profilegenerations within the same set of given input data constraints.

3. Yearly load profile The yearly load profile contains unique daily load profiles. Forinstance, the load profile constructed for day 1 can be quite different from that of,say, day 200 in a given year. This is a welcome change from the yearly load profilesthat are simply the same daily load repeated 365 times. This is helpful in sizingan off-grid system and understanding the system’s battery behaviour over a largerspan of time.

4. Adaptability The procedure described is flexible enough to incorporate more ordifferent appliances in the future or even change the specifications of any load thatmakes up the load profile. The load profiles can be further customized based onprecise demographic inputs and requirements from a specific community throughfieldwork, for instance.

5. Ease of system design Due to increasing needs of the off-grid population, SHS de-signers often need to oversize their systems to enable future growth of consumption.This could be planned better with the methodology which helps quantify the energyconsumption in the form of load profiles.

3.4. RESULTS AND DISCUSSIONSBased on the inputs described in Table 3.2 to the methodology described in Figure 3.5,different load profiles are obtained for the various tiers of the MTF. For ease of discussion,only the daily load profiles are shown here and discussed.

All the data from the generated load profiles have been made freely accessible at DOI:10.4121/uuid:c8efa325-87fe-4125-961e-9f2684cd2086.

3.4.1. STOCHASTIC LOAD PROFILES FOR MTFFigure 3.6 shows a tier 1 load profile for a representative day from the year-long loadprofile. As only a handful of loads operate in this case, both the peak power and the energyconsumption is the lowest compared to the rest of tiers.

Figure 3.7 shows a daily tier 2 load profile for a representative day. Similar to tier 1load profile, the characteristic peak occurs in the evening with moderate consumptionduring the day.

Figure 3.8 shows a daily tier 3 load profile for a representative day. An additional baseload can be seen in this case due to the fridge. As described earlier, the consumption ofthe fridge is not modelled based on the duty cycle. Instead, a base average consumptionis assumed from the manufacturer’s data, and the external usage events are modelled.

3

60 3. LOAD PROFILE CONSTRUCTION

0 4 8 12 16 20 240

2

4

6

8

10

12

Time [h]

Pow

er[W

]

Figure 3.6: Load profile of an off-grid household with tier 1 electricity access for a representative day. Totalenergy consumption for this day is 60 Wh with a peak of 12 W.

0 4 8 12 16 20 240

10

20

30

40

50

Time [h]

Pow

er[W

]

Figure 3.7: Load profile of an off-grid household with tier 2 electricity access for a representative day. Totalenergy consumption for this day is 262 Wh with a peak of around 51 W.

Figure 3.9 shows a daily tier 4 load profile for a representative day. Due to the highpower appliances used in this category, the peak power (around 1 kW) is much morepronounced.

Figure 3.10 shows a daily tier 5 load profile for a representative day. As this tier isassumed to contain appliances that support PUE as well, a very high peak of 2.5 kW canbe seen. Moreover, the peaks of all the load profiles lie in the peak window based on the

3.4. RESULTS AND DISCUSSIONS

3

61

0 4 8 12 16 20 240

30

60

90

120

150

Time [h]

Pow

er[W

]

Figure 3.8: Load profile of an off-grid household with tier 3 electricity access for a representative day. Totalenergy consumption for this day is 914 Wh with a peak of around 150 W.

0 4 8 12 16 20 240

200

400

600

800

1,000

Time [h]

Pow

er[W

]

Figure 3.9: Load profile of an off-grid household with tier 4 electricity access for a representative day. Totalenergy consumption for this day is 3.29 kWh with a peak of around 1 kW.

coincidence factor as explained in Section 3.3.2.

3.4.2. LOAD PROFILES: MAIN PARAMETERS

The main parameters of the load profiles discussed in Section 3.2.2 have been captured inTable 3.4.

Here, Pmax is the maximum value attained by PL,Peak throughout the year, while Pmin

3

62 3. LOAD PROFILE CONSTRUCTION

0 4 8 12 16 20 240

500

1,000

1,500

2,000

2,500

Time [h]

Pow

er[W

]

Figure 3.10: Load profile of an off-grid household with tier 5 electricity access supporting PUE loads for arepresentative day. Total energy consumption for this day is 9.34 kWh.

Table 3.4: Maximum peak power Pmax, minimum peak power Pmin, average daily energy Edaily and load factorfor the load profiles for each tier based on the 1-year generated load profile.

Tier 1 Tier 2 Tier 3 Tier 4 Tier 5

Pmax (W) 12 51 154 1670 3081Pmin (W) 6 35 113 583 1732Edaily (Wh) 50 218 981 3952 9531Load factor (-) 0.17 0.18 0.26 0.10 0.13

is the minimum value. When compared with the energy and power limits mentioned bythe MTF (Table 2.1), it can be seen that the constructed load profiles (in terms of Edaily

and Pmax) conform to these limits. The only exception is tier 3, where the Pmax is lowerthan the minimum limit as mentioned in the MTF. This is because of the super-efficientappliances already available and in use, which impact the overall power consumption.The impact of the efficiency of these appliances is more apparent in tier 3 than the lowertiers. The higher tiers 4 and 5 are still seeing advances being made in terms of dedicated,high power, efficient, off-grid appliances.

It must be noted that the existing solar lanterns, SHS or mini-grid solutions onlyspan tiers 1 to 3 in terms of off-grid electrification. There is still a long way to go beforemost current off-grid communities can reach tier 4 and 5 consumptions. Nonetheless,climbing up the rural electrification ladder is inevitable, and when the expected tier 4 and5 consumptions are reached, viable energy solutions need to be in place.

The higher tier load profiles can be seen to have lower load factors due to the tallpeaks of the high power appliances. This would have severe implications on the batterysize and the battery lifetime of standalone systems like SHS if that were to solely satisfy

3.4. RESULTS AND DISCUSSIONS

3

63

these load profiles.The values obtained in Table 3.4 can be tailored based on the kind of loads being

used and the corresponding usage restrictions that are captured in Table 3.2, which is theexpected use of this methodology.

3.4.3. IMPLICATIONS ON SYSTEM DESIGNThis section outlines the implications of the tier-based load profiles on the system designfor the tiers. A standalone solar PV and battery-based (off-grid) energy system is consid-ered in a location with around 4.5 equivalent sun hours (ESH), which is usual for mosttropical and equatorial regions.

Tier 1 load profile shown in Figure 3.6 could be typically satisfied with a 15 to 20 WpPV module and corresponding battery storage. This is already being done in the field bypico-solar products or smaller SHS designs. The tier 2 load demand shown in Figure 3.7could be typically satisfied with around 50 Wp PV module and corresponding batterysize for a location in the tropical belt, which is what a present-day small SHS can provide.Finding the optimal battery size requirement would need a more detailed analysis basedon the year-long load profile and the specific meteorological data.

Tier 3 load profile would need around 250 Wp of PV, which is about the typical size ofa contemporary residential solar panel used in the developed world. Only a handful SHSproviders currently operate with such PV dimensions when catering to household level,standalone, off-grid SHS. Tier 4 load profile would need almost 1 kWp of PV module. Aload peak of around 1 kW (Figure 3.9) and overall peak of 1.67 kW (Table 3.4 would haveimplications on the optimal dimensioning of the battery and the power electronics of thesystem, which needs a dedicated analysis.

Tier 5 load profile shows a load peak of around 2.5 kW (Figure 3.10) and an overallpeak of around 3 kW due to the high power appliances that enable PUE. A standalonesystem would demand a PV of around 2.5 kWp and corresponding battery and powerelectronics to match. This shows that supporting the high power, dedicated, PUE-enablingappliances is going to be an exacting challenge on the size of a standalone system likeSHS. Guaranteeing zero loss-of-load events would call for a highly oversized standalonesystem. A microgrid with distributed generation might be a more suited option 1.

3.4.4. COMPARISON WITH FIELD DATAAt the time of conducting the work described in this chapter, most prevalent SHS andother off-grid systems deployed at household level were only catering up to tier 2, or insome cases, tier 3 level electricity access. BBOXX, an SHS provider in Rwanda, East Africaoffers a portfolio of efficient DC appliances along with the SHS. Figure 3.11 shows a singleday load profile of an off-grid SHS from BBOXX capable of powering 3 LED lights (1.2W nominal consumption) and a USB port for charging a mobile phone and a portableradio. The consumption limits would deem this to be a tier 1 load profile. Also shown is astochastically generated load profile, with different inputs matching those correspondingto the appliances of the SHS in the field. The generated profile seems to match themeasured one closely, especially with respect to the peak load in the peak window in

1Note that these system sizes are back-of-the-envelope approximations, and a thorough analysis is needed tocomment on the system sizes that satisfy particular system metrics, which is exclusively dealt with in Chapter 5

3

64 3. LOAD PROFILE CONSTRUCTION

0 4 8 12 16 20 240

2

4

6

8

Time [h]

Pow

er[W

]

measuredgenerated

Figure 3.11: Comparison of a generated stochastic load profile with the measured load consumption over asingle day of a household in Rwanda powered by an SHS from BBOXX.

the evening. The total daily energy of the measured load profile is 35.7 Wh, while thatof the generated one is 34.7 Wh, representing a 2.9% error. Furthermore, the load factorof the measured load profile is 0.18 while that of the generated load profile is 0.17. Amuch greater match can naturally be achieved if needed by adjusting the operationalconstraints. However, the comparison here is to merely exemplify the usefulness of thedescribed methodology.

3.5. CONCLUSIONSThis chapter described a bottom-up, stochastic load profile construction methodology toquantify the energy needs for the various tiers of the MTF. The loads were entirely com-posed of dedicated, off-grid, and in some cases the so-called super-efficient, appliances.Advantages like scalability, adaptability, and randomness of the proposed methodologywere identified. A method for incorporating coincidence factor as a design input for themethodology was also introduced. Several stochastic load profiles were created using thedescribed methodology, and their impact on standalone system design for the differentload profiles has been underlined. The impact of appliances enabling productive useof energy was also investigated through a tier 5 load profile construction. The utilityof this methodology can be greatly augmented with the availability of local data andcomplemented with targeted surveys per target community/region.

The load profile construction methodology described in this chapter is expected togreatly help various off-grid electrical system designers in constructing load profilesand customizing energy solutions to cater to the growing energy needs of the un(der)-electrified population.

In the next chapter, estimating battery lifetime is discussed for SHS-specific applica-tions utilizing the load profiles constructed through the methodology described in this

REFERENCES

3

65

chapter.

RECOMMENDATIONS AND FUTURE WORKThe methodology described assumes rated power consumption of these DC appliancesthroughout the usage of the appliance. In reality, some appliances may consume differ-ently based on their usage. For now, only the fridge has been treated as a special case.Other examples could include the starting power consumption of an appliance, or thevarying power consumption of an LCD TV depending on the screen illumination, whichcan be quite different from the rated power. In the absence of actual power consumptionprofiles of these up-and-coming DC appliances, the current load profile constructionmethodology is considered to be sufficient. More light will be hopefully shed in the futureby the real-time data gathered from SHS in the field. Seasonal implications on the loadvary for different locations, and were not considered in this study. However, these canalso be added as an additional parameter based on location specific information; theinput table (Table 3.2) can be updated based on context specific knowledge, potentiallycomplemented by field studies.

ACKNOWLEDGEMENTThe authors are grateful to Ashley Grealish from BBOXX for providing the 1-day usagedata from their SHS in Rwanda.

REFERENCES[1] B. Tenenbaum, C. Greacen, T. Siyambalapitiya, and J. Knuckles, From the bottom up:

how small power producers and mini-grids can deliver electrification and renewableenergy in Africa. World Bank Publications, 2014.

[2] C. Wolfram, O. Shelef, and P. Gertler, “How Will Energy Demand Develop in theDeveloping World?,” Journal of Economic Perspectives, vol. 26, no. 1, pp. 119–138,2012.

[3] IEA, Energy Access Outlook 2017 - From Poverty to Prosperity. Organization forEconomic Cooperation and Development, International Energy Agency, 1 ed., 2017.

[4] A. A. Phadke, A. Jacobson, W. Y. Park, G. R. Lee, P. Alstone, and A. Khare, “Powering ahome with just 25 watts of solar pv. super-efficient appliances can enable expandedoff-grid energy service using small solar power systems,” tech. rep., Lawrence Berke-ley National Laboratory (LBNL), Berkeley, CA (United States), 2015.

[5] Global LEAP, “The State of the Off-Grid Appliance Market,” tech. rep., Tech. rep.Washington, DC: Global LEAP Lighting and Energy Access Partnership, 2016.

[6] W. Y. Park and A. A. Phadke, “Adoption of energy-efficient televisions for expandedoff-grid electricity service,” Development Engineering, vol. 2, pp. 107 – 113, 2017.

[7] REN21, Renewables 2017 Global Status Report. Paris, REN21 Secretariat, 2017.

3

66 REFERENCES

[8] GIZ, Photovoltaics for Productive Use Applications - A Catalogue of DC-Appliances.Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, 2016.

[9] A. L. Kooijman-van Dijk, The Power to Produce: The role of energy in poverty reductionthrough small scale Enterprises in the Indian Himalayas. University of Twente, 2008.

[10] B. Troja, “A quantitative and qualitative analysis of the super-efficient equipmentprogram subsidy in India,” Energy Efficiency, vol. 9, no. 6, pp. 1385–1404, 2016.

[11] A. H. Smets, K. Jäger, O. Isabella, R. Van Swaaij, and M. Zeman, Solar Energy, thePhysics and Engineering of Photovoltaic Conversion Technologies and Systems. UITCambridge, 2016.

[12] S. Treado, “The effect of electric load profiles on the performance of off-grid resi-dential hybrid renewable energy systems,” Energies, vol. 8, no. 10, pp. 11120–11138,2015.

[13] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrifica-tion,” in 2017 International Conference on the Domestic Use of Energy (DUE), pp. 8–15,April 2017.

[14] M. Gustavsson, “With time comes increased loads: An analysis of solar home systemuse in lundazi, zambia,” Renewable Energy, vol. 32, no. 5, pp. 796–813, 2007.

[15] S. Hirmer and P. Guthrie, “The benefits of energy appliances in the off-grid energysector based on seven o ff -grid initiatives in rural Uganda,” Renewable and Sustain-able Energy Reviews, vol. 79, no. May, pp. 924–934, 2017.

[16] F. Riva, H. Ahlborg, E. Hartvigsson, S. Pachauri, and E. Colombo, “Electricity ac-cess and rural development: Review of complex socio-economic dynamics andcasual diagrams for more appropriate energy modelling,” Energy for SustainableDevelopment, vol. 43, pp. 203–223, 2018.

[17] F. Riva, A. Tognollo, F. Gardumi, and E. Colombo, “Long-term energy planning anddemand forecast in rural areas of developing countries: classification of case studiesand insights for a modelling perspective,” Energy Strategy Reviews, vol. 20, pp. 71–89,2017.

[18] C. Blodgett, P. Dauenhauer, H. Louie, and L. Kickham, “Accuracy of energy-usesurveys in predicting rural mini-grid user consumption,” Energy for SustainableDevelopment, vol. 41, pp. 88–105, 2017.

[19] E. Hartvigsson and E. O. Ahlgren, “Comparison of load profiles in a mini-grid: Assess-ment of performance metrics using measured and interview-based data,” Energy forSustainable Development, vol. 43, pp. 186–195, 2018.

REFERENCES

3

67

[20] S. C. Bhattacharyya and G. R. Timilsina, “Modelling energy demand of developingcountries: Are the specific features adequately captured?,” Energy Policy, vol. 38,no. 4, pp. 1979–1990, 2010.

[21] L. G. Swan and V. I. Ugursal, “Modeling of end-use energy consumption in theresidential sector: A review of modeling techniques,” Renewable and SustainableEnergy Reviews, vol. 13, no. 8, pp. 1819–1835, 2009.

[22] G. Tsekouras, P. Kotoulas, C. Tsirekis, E. Dialynas, and N. Hatziargyriou, “A patternrecognition methodology for evaluation of load profiles and typical days of largeelectricity customers,” Electric Power Systems Research, vol. 78, no. 9, pp. 1494 – 1510,2008.

[23] T. Räsänen, D. Voukantsis, H. Niska, K. Karatzas, and M. Kolehmainen, “Data-basedmethod for creating electricity use load profiles using large amount of customer-specific hourly measured electricity use data,” Applied Energy, vol. 87, pp. 3538–3545,2010.

[24] J. Nuno, M. António, and L. Ribeiro, “A new clustering algorithm for load profilingbased on billing data,” Electric Power Systems Research, vol. 82, no. 1, pp. 27 – 33,2012.

[25] I. A. Sajjad, G. Chicco, and R. Napoli, “A probabilistic approach to study the load vari-ations in aggregated residential load patterns,” in 2014 Power Systems ComputationConference, pp. 1–7, Aug 2014.

[26] P. Boait, V. Advani, and R. Gammon, “Estimation of demand diversity and dailydemand profile for off-grid electrification in developing countries,” Energy for Sus-tainable Development, vol. 29, pp. 135–141, 2015.

[27] J. V. Paatero and P. D. Lund, “A model for generating household electricity loadprofiles,” International Journal of Energy Research, vol. 30, no. 5, pp. 273–290, 2006.

[28] S. Mandelli, M. Merlo, and E. Colombo, “Novel procedure to formulate load profilesfor off-grid rural areas,” Energy for Sustainable Development, vol. 31, pp. 130–142,2016.

[29] H. Louie and P. Dauenhauer, “Effects of load estimation error on small-scale off-gridphotovoltaic system design , cost and reliability,” Energy for Sustainable Develop-ment, vol. 34, pp. 30–43, 2016.

[30] Global LEAP, “Off-grid appliance market survey,” tech. rep., Tech. rep. Washington,DC: Global LEAP Lighting and Energy Access Partnership, 2015.

[31] SunDanzer, “DCR50 DC powered refrigerator,” Datasheet, 2016.

[32] Phocos, “Led Light SL 1220CF120,” Datasheet, 2016.

[33] Samsung, “Mobile phone Guru Plus,” Datasheet, 2016.

3

68 REFERENCES

[34] Global LEAP, “Global leap awards - 2016 buyer’s guide for outstanding off-grid fansand televisions,” tech. rep., Tech. rep. Washington, DC: Global LEAP Lighting andEnergy Access Partnership, 2016.

[35] HP, “HP Pro Tablet EE G1,” Datasheet, 2016.

[36] Panasonic, “Automatic Rice Cooker SR-3NA-5,” Datasheet, 2016.

[37] Hotspot Energy, “Solar/DC Air Conditioner DC4812VRF,” Datasheet, 2016.

APPENDIXTable 3.5 lists down the loads used in the study. The model number and basic spec-ifications of the loads are mentioned, along with the corresponding source whereverapplicable.

Table 3.5: List of appliances used for constructing the load profile.

Loads Sample model number Source

LED Lighting SL 1220CF120 [32]Mobile Phone Samsung Guru Plus [33]Radio Fosera, FS106 [8]Fan ONergy 10" BOX FAN, FS91, ONergy 16"

PEDESTAL FAN, FS92, Ceiling Fan ME-103-DC[8], [34]

TV Fosera DC TV 15.6” 12 V, D.light design 18.5"LE185N91C,Mobisol 24" MSDV2310MY-308C1

[34]

Fridge Solar Chill, FS52 , Sundanzer DCR 50, DCR 165 [31],[8]Tablet HP Pro Tablet 10 EE G1 [35]Kettle Solar DC Kettle SE520, FS65 [8]Laptop Generic Laptop -Rice Cooker SR-3NA-S [36]Clothes Iron Solar DC Power Iron Dry/ Spray style-12V SL100S, FS192 [8]Washing Machine Washing Machine CERAD, FS127 [8]Air cooler DC Solar Air Conditioner, DC4812VRF [37]Power tools Bosch 18V Lithium Ion 4-Tool Combo Kit (CLPK414-181), FS84 [8]Grinders/Millers Grain Mill Solar Milling, FS32 [8]Sewing Machine Sewing Machine CERAD, FS73 [8]Water Pump Solar Surface Slow Pump Dankoff, FS10 [8]

4ESTIMATING BATTERY LIFETIME IN

SOLAR HOME SYSTEMS

It has become appallingly obvious that our technology has exceeded our humanity

Albert Einstein

ABSTRACTThe battery is a vital but usually the most expensive part of an SHS. As the battery has theleast lifetime among other SHS components, it is also the first to fail. Estimating batterylifetime is a critical task for SHS design. However, it is also a complex task due to thereliance on experimental data or modelling cell level electrochemical phenomena forspecific battery technologies and application use-case. This chapter presents a practical,non-empirical battery lifetime estimation methodology specific to the application andthe available candidate battery choices. An application-specific SHS simulation is carriedout, and the battery activity is analyzed. A practical dynamic battery lifetime estimationmethod is introduced, which captures the fading capacity of the battery dynamicallythrough every micro-cycle. This method is compared with an overall non-empiricalbattery lifetime estimation method, and the dynamic lifetime estimation method is foundto be more conservative but practical. Cyclic ageing of the battery is thus quantified andthe relative lifetimes of 4 battery technologies are compared, viz. Lead-acid gel, Flooded

This chapter is based on the following publications:

1. N. Narayan,T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, & M. Zeman. (2018).Estimating battery lifetimes in Solar Home System design using a practical modelling methodology.Applied Energy, 228, 1629-1639.

2. N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P. Bauer and M. Zeman, "A simple method-ology for estimating battery lifetimes in Solar Home System design," 2017 IEEE AFRICON, Cape Town,2017, pp. 1195-1201. doi: 10.1109/AFRCON.2017.8095652.

69

4

70 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

lead-acid, Nickel-Cadmium (NiCd), and Lithium Iron Phosphate (LiFePO4) battery. Forthe same SHS use-case, State-of-Health (SOH) estimations from an empirical modelfor LiFePO4 is compared with those obtained from the described methodology, and theresults are found to be within 2.8%. Based on the intended application and batterymanufacturer’s data, the practical methodology described in this chapter can potentiallyhelp SHS designers in estimating battery lifetimes and therefore making optimal SHSdesign choices.

OUTLINEThis chapter is divided into five sections. Section 4.1 introduces the work described inthe chapter, while Section 4.2 outlines the necessary technical background and the focusof this study. Section 4.3 discusses the specific application use-case and describes themethodology used. Section 4.4 discusses the results and compares the estimated batterylifetime from the proposed methodology with an experimentally obtained model forLiFePO4 battery. Finally, Section 4.5 concludes the discussion on the battery lifetimeestimation methodology.

4.1. INTRODUCTIONThe battery is a vital component of the SHS, enabling energy storage of the PV output.However, the sizable proportion of upfront battery costs makes the battery the mostexpensive SHS component, as seen in Figure 4.1. This fact, coupled with the low batterylifetimes (sometimes as low as 3 years [1]), makes battery-costs the most relevant in SHSdesign. Battery-costs recur not just in terms of the upfront costs, but also in terms of thereplacements during SHS lifetime, thereby making battery lifetime a critical parameter inSHS applications.

200 300 400 500 600

SHS

Total Retail Costs [$]

PV Battery BOS Appliance

Figure 4.1: Split-up of upfront costs of SHS components for powering a 19" TV, radio, lights and a mobile phonecharger in 2014 (data from [2]).

Additionally, the accurate sizing of the battery can impact the battery lifetime [3]. Thisis because an increase in battery size for the same application reduces the average Depthof Discharge (DOD) for the battery [4], as seen below in Section 4.3. Therefore, the sizingof the battery in an SHS presents itself as an interesting balance between upfront costs(size) and lifetime [5].

Battery lifetime is thus seen as a vital parameter to be considered in SHS design,especially in the cost-sensitive context of electrification.

4.1. INTRODUCTION

4

71

4.1.1. LITERATURE STUDYBattery lifetime estimation models can be broadly classified under two categories, viz.performance-based models and cycle counting models.

In performance-based models, the performance values of the battery are simulatedbased on certain performance parameters. When the particular parameter drops belowa pre-determined value, the end of life (EOL) is reached for the battery. These can beclassified into 4 different categories.

(i) Electrochemical Models that need extensive information on the chemical andphysical interactions occurring within the battery in order to accurately model thebattery. Despite the efforts to simplify these models while effectively estimating thedynamically changing states (SOC and SOH) [6], the lack of detailed informationsuch as Li-ion concentration, diffusion coefficients, reaction rates, ionic conduc-tivity, reduce the applicability of these models specially when different batterytechnologies are to be compared.

(ii) Equivalent Electrical Circuit Models that represent the battery as an equivalentelectrical circuit comprising electrical elements like resistors, capacitors, voltage,and current sources. These kind of models can also include the thermal stressesrelated to fast charging to develop optimal charging profiles [7]. A combinedelectrochemical-thermal model has also helped in recent studies to develop health-aware strategies for fast charging of Li-ion battery, underlining the importance ofbeing able to model SOH in battery applications [8].

(iii) Analytical Models with empirical data fitting are constructed by interpolating andfitting empirical data obtained through experiments.

(iv) Artificial neural networks (ANN) can establish a relationship between the systemoutputs and input operating conditions, given a large enough dataset [9].

Performance-based models can rely on experimental methods that observe and mea-sure the battery degradation through time, or on semi-empirical approaches that repre-sent the fade mechanisms using equations that fit a particular type of cell. These modelscan also be constructed by including the physiochemical effects behind the side reac-tions of a particular set of chemical compounds endemic to particular cell chemistries.Therefore, the main limitation of all of the underlying approaches for performance-based models is that they are constructed for specific cells, under certain environmentconditions, and tested through a limited amount of time [10]. Additionally, some semi-empirical approaches do not consider the effect of parameters like DOD [11]. For modelsrelying on ANN, a vast amount of data is first needed for the neural network to be trainedreliably. In general, it can be said that to construct reliable performance-based models,it not only costs time per specific intended application and use-case needed but alsoraises concerns on the accuracy of these approaches if they were to be used under adifferent set of conditions wherein the same battery technology undergoes a differentstress pattern .

In this PhD project, it was also endeavoured to create dynamic performance models of lead-acid gel andLiFePO4 battery types for SHS applications. To this end, cell level experiments were conducted to constructequivalent electrical circuit models. This is described in Appendix 8.4.3.

4

72 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

Unlike performance-based models, cycle counting or weighted Ah-throughput (chargeprocessed by the battery) models are able to determine parameters which can be linkedto their EOL, for instance, Ah-throughput (the amount of energy processed by the bat-tery), cycles, or time since manufacturing [9]. A comparison study in the past has shownaccurate results for lifetime prediction with weighted Ah-throughput model [12]. Thesemodels are mainly based on the data provided by the manufacturers, assuming thatthe battery is able to achieve an overall Ah-throughput throughout its life under certainspecific stress factors like DOD and temperature. Palmgren-Miner (PM) rule is one of themost common examples that fit in the category of cycle counting models. The differentlifetime estimation models are compared in Table 4.1.

Table 4.1: Comparison of the performance-based and cycle-counting models in the context of battery lifeestimation. Adapted from [9].

Model type Model exam-ple

Merits Demerits

Performance-based

Electro-chemical

Very high accuracy Complex

Low-speed

EquivalentElectricalCircuit Model

Good prediction of dy-namic behaviour

Depends on exper-imental data foraccuracyNot suitable for lifetimeprediction as is

Analyticalmodel withempiricaldata fitting

Ease of simulation us-ing the model due tothe analytical functions

Effects of various stressfactors need to be com-bined from the experi-mental data

Good accuracy Large amount of dataneeded

Artificial Neu-ral Networks(ANN)

Reasonable accuracywith high speed

Adequate training ofANN requires a largedataset

Knowledge of bat-tery mechanisms notneeded

Cycle-counting

Palmgren-Miner (PM)rule

Deviations from stan-dard operating condi-tions captured well

Although non-empirical in itself,relies on empirical datafrom the manufacturer

Simple implementa-tion

There have been a few cycle-counting-based lifetime estimation models discussedin the past in [13, 14, 15], and most notably in [5], where the authors discuss a simple

4.2. BATTERY LIFETIME

4

73

methodology for non-empirical battery lifetime estimation. However, while [13] assumesan arbitrary duration of a cycling event 6400 seconds long, [14] uses only temperatureas a stress factor but not for battery storage specifically, [15] employs off-line cycle-counting through rainflow counting algorithm but using DOD as the only stress factorfor battery degradation, and [5] does not take into account dynamic capacity fadingand temperature as an additional stress factor. The work described in [16] presents alifetime estimation of lead-acid battery, but also does not consider dynamic capacityfading, and the capacity loss is only estimated after one entire year of simulation usingrainflow counting algorithm.

This chapter focuses on a non-empirical, cycle-counting approach for the estimationof the battery lifetime while also taking into account important battery stress factors likeDOD and temperature. Given the off-grid SHS application, where the typical batteryC-rates are C/20 to C/10 [17, 18], the C-rates are considered to be low enough to not beincluded as a critical stress factor. The dynamic capacity fading model proposed here iscapable of estimating the State of Health (SOH) after every micro-cycle-based degradationwhile the battery is under operation. Additionally, multiple technologies are used in thisstudy, demonstrating the usefulness of the proposed methodology across multiple batterytechnologies without having to model the electrochemical processes at the cell level forthe different technologies.

4.1.2. CONTRIBUTIONS OF THIS CHAPTERFollowing are the main contributions of this chapter.

• A methodology to estimate the battery lifetime in a practical way using the application-based, expected battery usage and the battery data provided by the manufacturer.The methodology is applicable across battery technologies.

• Insight into performance of 4 different battery technologies for the SHS application.

• A dynamic capacity fading model that quantifies the fractional degradation causedby the micro-cycles of battery activity.

4.2. BATTERY LIFETIME

4.2.1. BATTERY PARAMETERSSome of the important battery parameters that this work refers to are discussed in thissection.

State of Charge and Depth of Discharge The state of charge (SOC) indicates batterycharge as a fraction of the initial capacity, while the depth of discharge (DOD) refers to thecapacity discharged as a fraction of the initial capacity. They are treated as a complementof each other.

The DOD can be calculated from the battery discharge current over a discharginginterval as shown in Equation 4.1.

DOD =∫ tf

tiIdischarge.d t

Ci(4.1)

4

74 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

where tf is the time at the end of the discharge interval process, ti is the initial time,Idischarge is the current, and Ci the initial battery capacity.

Cycle-life Cycle-life is defined as the number of charge/discharge cycles that a batterycan undergo maintaining a specified percentage of its initial capacity, after which batteriesreach the end of life (EOL). The EOL is usually defined as 80% of the initial rated batterycapacity [19].

State of Health (SOH) State of Health refers to the fraction of the rated battery capacityactually available for cycling. As the battery ages, the SOH reduces, and 80% SOH is oftendefined as the EOL for a battery. The battery ageing can be classified into two categories,viz., cyclic ageing and calendar ageing.

Cyclic ageing Cyclic ageing is related to the decrease in battery capacity while thebattery is undergoing cycling. Cyclic ageing plays a larger role in cases where operationtimes are relatively longer than idle periods, like in off-grid SHS.

Calendar ageing Calendar ageing is related to the decrease in battery capacity while thebattery is under storage and not in use, and therefore, independent of charge/dischargecycles. Calendar ageing plays a predominant role in cases where operation times areshorter than idle periods [20]. This work takes into account only the cyclic ageing process.In other words, this work models the application-specific, battery induced capacity fadingand the resulting battery lifetime.

Active SOC/DOD Active SOC/DOD refers to the State-of-Charge or Depth-of-Dischargeof the battery while the battery is in operation [5]. The concept of active SOC/DOD helpsin looking at only those SOC/DOD points that actively contribute towards cyclic ageing ofa battery, which is the focus of this work.

4.2.2. CAUSES OF BATTERY DEGRADATIONThe loss of active lithium, principally at the anode, is the main reason for lithium-ion celldegradation. The mechanical stresses induced by cycling and temperature gradients leadto contraction and expansion, eventually increasing the cell impedance and reducing thecell capacity [20]. Electrode pore clogging, passive layer growth, and lithium metal platingare consequences of the undesired side reaction that takes place inside the Lithium-ion cells [21]. In the case of lead-acid batteries, the major degradation processes thatdecrease performance and provoke end of service life are anodic corrosion, active massdegradation, and loss of adherence to the grid plate [22]. Furthermore, the continuousloss of water in vented cells and the production of lead sulphate (in the electrodes) out ofthe active materials of the cell facilitates the ageing process [23]. NiCd cells suffer fromhydrogen production when overcharged and when exposed to a temperature higher thannominal temperature; therefore, the accurate detection of the end-of-charge processis fundamental to avoid battery damage [24]. In NiCd cells, Cd crystallization happens

4.3. METHODOLOGY

4

75

under particular cycling and storage conditions, causing a reduction of active materialwhile reducing the active surface dedicated to the electrochemical reaction [19].

Battery lifetime is directly related to battery usage, and therefore the way in whichthe battery cycling occurs and the environmental conditions affect the natural ageing ofbatteries. One of the principal factors to take into account is the DOD range of operation,which defines the usable capacity of the battery, and therefore, the battery range ofoperation. Another parameter that profoundly impacts battery ageing is cell temperature,where an increase in temperature translates into a higher electrochemical activity thatinstantaneously improves cell performance but in the long term fosters side reactionsthat consume the active materials, thereby diminishing battery capacity [25]. The kineticsof the electrochemical reaction also relates to the charge and discharge rates imposed bythe applications. Therefore, the methodology introduced in the next section takes intoaccount the influence of temperature and cycling to quantify their impact on lifetime.

4.3. METHODOLOGYThe methodology followed in this study is composed of multiple steps. Section 4.3.1describes the extraction of lifetime data from the battery datasheet as a function of tem-perature and DOD. Section 4.3.2 details the SHS-based inputs used in the methodology.Sections 4.3.3 and 4.3.4 discuss the two models that are proposed and used in this study,viz. the overall battery usage model and the dynamic capacity fading model.

4.3.1. BATTERY DATA FROM THE MANUFACTURERBattery manufacturers usually provide the battery cycle-life characteristics as a functionof DOD and temperature, for example, in [26]. The data from these curves are extracted,and lookup-functions are created.

Battery technology Four different battery technologies are chosen for this study, viz.,sealed Lead-Acid or Lead-Acid gel (LA-gel), Lithium-Iron Phosphate (LiFePO4), floodedLead-Acid (LA), and Nickel-Cadmium (NiCd) battery. The corresponding datasheetsused are [26, 27, 28, 29] respectively. While the merits and demerits of using one batterytechnology over the other for SHS application is a different subject of discussion alto-gether, the reason for including these different technologies in this study is to illustratethe usefulness of the described methodology to estimate the battery lifetime irrespectiveof the underlying battery chemistry.

Temperature linearity For cycling at a given DOD level, it is observed that the cycle lifeshows a linear temperature dependency, at least in the range of 20 to 45C, a temperaturerange typically mentioned in the datasheets. This behaviour is illustrated in Figure 4.2 forthe flooded lead-acid battery.

Polynomial approximations The linearity on temperature dependency is exploited tocreate polynomial approximation functions. A 4th order polynomial approximation forthe battery lifetime curves for the above-mentioned technologies is given by the followingset of Equations.

4

76 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

20 25 30 35 40 450.4

0.6

0.8

1

Temperature [C]

No

rmal

ized

cycl

eli

fe[-

]

Figure 4.2: Normalized cycle life with temperature for a flooded lead-acid battery at 30% DOD with a referencetemperature of 20C. Data sourced from [28]).

n(T,DOD) = n(Tref,DOD)− f (Tavg)Dn(DOD) (4.2)

n(Tref) = p4d 4 +p3d 3 +p2d 2 +p1d +p0 (4.3)

f = pl1 Tavg +pl0 (4.4)

Dn = pd4 d 4 +pd3 d 3 +pd2 d 2 +pd1 d +pd0 (4.5)

Where

n(T,DOD) = Cycle life for a given temperature and DOD

f = A linear factor

Dn = Difference of n between two temperatures

p0 to p4 = Polynomial fitting coefficients at Tr e f

pl1 , pl0 = Fitting coefficients for determining

the linear factor

pd0 to pd4 = Fitting coefficients for difference

between two temperature curves

Tavg = Average operating temperature

Tref = Reference operating temperature

d = Battery DOD

A reconstruction of the battery lifetime curves depending on temperature and DODfrom the battery datasheet is shown in Figure 4.3 for a flooded lead-acid battery. These

4.3. METHODOLOGY

4

77

curves are then approximated as polynomial ‘lookup’ functions that can be used in batterylifetime estimation based on the application-specific battery usage.

20 30 40 50 60 70 800

2,000

4,000

6,000

8,000

DOD [%]

Cyc

leli

fe[-

]T=20CT=25CT=30CT=35CT=40CT=45C

Figure 4.3: Reconstruction of battery cycle-life curves based on DOD for flooded lead-acid battery(data sourcedfrom [28]).

It must be noted that many manufacturers do not explicitly distinguish betweenthe ambient temperature and the actual battery operating temperature, and some evenmention the battery cycle-life curves at ambient temperature [28, 26]. It is difficult toaccurately determine the battery temperature, as it depends on several local operatingconditions outside the electrical functioning of the system. In this study, the batteryoperating temperatures are taken to be at ambient levels, especially given the slow C-ratesfor the SHS batteries.

4.3.2. SHS APPLICATION AND LOAD PROFILEAn SHS application was considered as the use-case for investigating the battery usageand estimating the battery lifetime. The various inputs used for the SHS use-case for aredescribed below.

LOAD PROFILE

As discussed before in Chapter 3, load profiles were constructed for the various tiers ofthe multi-tier framework based on the stochastic load profile construction methodology.For the SHS use-case described in this chapter, a tier 3 load profile was used, in line withthe consumption limits outlined by the MTF [30]. This was done taking into accountthe efficient DC appliances that are on the rise in the off-grid market. Several works inthe recent past have indicated a sharp rise in the availability as well as popularity of theso-called super-efficient DC appliances [2, 31, 32]. The operational (DC) loads that makeup the load profile are LED lights, mobile phone charging, radio, TV, fridge, and a tablet,as described in Chapter 3.

4

78 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

METEOROLOGICAL INPUTS

The meteorological inputs to the model were obtained from the tool Meteonorm [33],which includes irradiance, wind speed and temperature. A sample geographical locationexperiencing an average of 5.5 equivalent sun hours per day was considered. The dataresolution of the inputs to the model was 1-minute. Although far from instantaneous, it isassumed that a one-minute data resolution provides reasonable accuracy to incorporatethe intermittency of the PV generation and the load profile.

SHS PV SIZE

The selected PV module is Jinko Solar JKM265P rated 265 Wp. One PV module is enoughto cover the average daily energy needs for the given load profile. Additionally, the extraDC yield helps to compensate for the system inefficiencies. The PV output is modelledand corrected for thermal losses. As it is not the primary focus of this chapter, the detailedPV model is described in Appendix B.

SHS BATTERY SIZE

Loss of Load probability (LLP) was chosen as the optimizing parameter for finding thedesired battery size. LLP is a metric defined as the ratio of the expected amount ofdowntime (system failure) of the system while delivering the demanded power, to thetotal amount of time the system was designed to deliver power for. The concept of LLP hasbeen explained in a previous work in [4], and is explored in detail in Chapter 5, containingdetailed SHS-level modeling and simulation. A battery size of 1440 Wh was consideredfor the given load profile and the chosen solar panel. This PV-battery combinationguaranteed an LLP of 1.8% for the given load profile, i.e., this was the minimum storagesize with the required PV size that could guarantee a total loss of load of 1.8% throughoutthe 1 year of simulation. The LLP optimization approach for the given load profile can beseen in Figure 4.4, where the chosen battery size and the corresponding LLP are markedon the graph.

SHS SIMULATION

An SHS model was constructed in MATLAB to simulate the functioning of the SHS for ayear. Based on the load profile and the modelled PV output, the simulation was run overan arbitrary calendar year to obtain the battery usage pattern. It should be noted thatthe battery was limited to a maximum DOD limit of 80% in the simulation. The batteryefficiency was assumed to be constant throughout the simulation, and certain constantefficiencies were assumed after a comparative study based on literature [34, 35, 36, 37, 38].The efficiencies assumed are 85%, 90%, 78%, 93%, for flooded LA, LA-gel, NiCd, andLiFePO4 respectively. Additionally, a constant electronic power conversion efficiency of95% was assumed for the SHS. These efficiency numbers can also be specifically changeddepending on the data from the manufacturer or the choice of a particular battery for theapplication without impacting the efficacy of the described methodology.

4.3.3. OVERALL BATTERY USAGE MODELIn the overall battery usage method, the lifetime is estimated based on the overall bat-tery usage that is extracted from the 1-year SHS simulation. The main parameters for

4.3. METHODOLOGY

4

79

0 2 4 6 8 100

0.2

0.4

0.6

0.8

Battery size [kWh]

LLP

[-]

Figure 4.4: Loss of load probability (LLP) curve for the given load profile with varying storage sizes. A batterysize of 1440 Wh (marked on the graph) is deemed sufficient through this analysis, resulting in an LLP of 1.8%.

estimating the battery lifetime are the average DOD and average temperature in batteryoperation. These were derived from the battery usage in two different ways, viz. thecoarse average DOD and the micro-cycle zero-crossings-based approach.

COARSE AVERAGE APPROACH

This is the less computationally demanding approach of the two, where the requiredaverage DOD and temperature values are simply taken as the average of all the data pointsthroughout the simulation, as shown in Equation 4.6.

DOD =∑N

i=1 DODi

N(4.6)

T =∑N

i=1 Ti

N

Where DOD and T are the average DOD and temperature, N is the total number of datapoints in the simulation.

While computationally less intensive, a simple average does not capture the batteryusage well. This is because not only does this method include the extraneous, inactivebattery periods, but also the battery processes different amounts of energy throughputunder different DODs. Therefore, a more weighted approach needs to be used for takingthis into account.

ZERO-CROSSING (ZC) APPROACH

In the micro-cycle zero-crossing (ZC) approach, micro-cycles of battery activity are de-fined based on the zero-crossings of the battery current. Therefore, only the active battery

4

80 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

cycling periods are considered. This concept was introduced in a recent publication fromthe authors [5].

A micro-cycle in this context is defined as a small cycle of variable duration that existsbetween two consecutive current zero crossings; this is typically much shorter than a fullcharge-discharge cycle. This method involves taking into account all the zero currenttransitions given the 1 minute data resolution.

Figure 4.5 illustrates this concept. In this case, the active DOD should only be consid-ered for the durations that the battery is cycling. Therefore, the states with battery currentIbattery = 0 are ignored. Note that the concept can be equivalently explained if the batterypower Pbattery is considered instead of Ibattery. In that case, the area of each micro-cyclewould give the energy that the battery cycles in that interval.

i = 1 i = 2

i = 3

i = 4

Ibattery(t )

t

Di schar g i ng

C har g i ng

Figure 4.5: An illustrative battery current waveform showing arbitrary micro-cycle intervals, where i is themicro-cycle number.

Finally, with the micro-cycle zero-crossing data extracted from the battery simulation,the average active DOD can be calculated as shown in Equation 4.7.

DOD =

N∑i=1

DODi .Ethri

N∑i=1

Ethri

(4.7)

Where DOD : Combined average active DOD due to all the micro-cycles, DODi : Averageactive DOD in the i th micro-cycle, Ethri : Total energy throughput in the i th micro-cycle,N : total number of ZC-based micro-cycles.

The average temperature is calculated based on the duration of the ZCs as follows.

T =

N∑i=1

Ti .TZCi

N∑i=1

TZCi

(4.8)

Where T : Combined average temperature due to all the micro-cycles, Ti : Averagetemperature in the i th micro-cycle, TZCi : Total duration of the i th micro-cycle.

4.3. METHODOLOGY

4

81

LIFETIME ESTIMATION BASED ON OVERALL BATTERY USAGE

Once the DOD is known, the cycle-life number n is then obtained from the look-upfunctions described in Section 4.3.1. The battery lifetime is then estimated as shown inEquation 4.9 below [5].

L = n ×DOD × 2×Enom

N∑i=1

Ethri

(4.9)

Where (L) is the battery lifetime in years, and Enom is the nominal battery energy capacityin Wh.

Assumptions under overall battery usage methodology In the overall battery usagemethodology, there are a few crucial assumptions that serve the balance between thecomplexity of implementation and accuracy. These are as follows.

1. The battery performs throughout the 1-year simulation without any loss in capacity.

2. The battery performance in terms of the dynamic SOC is irrespective of its SOH.That is, the SOC of the battery is not corrected for its faded capacity during thesimulation.

3. Only one year of intended battery usage is enough to be able to linearly extrapolatethe battery usage and hence the cycle life.

The major drawback of this method is that there is no loss in capacity of the batteryduring the simulation. Instead, the battery lifetime is estimated based on the overallbattery usage for one year. It would be more accurate to estimate the battery lifetimewhile taking the capacity fading into account dynamically in the battery performance, asdiscussed in the next section.

4.3.4. DYNAMIC CAPACITY FADING MODELThe dynamic capacity model fundamentally differs from the overall battery usage methodfor battery lifetime estimation, in that this model dynamically assesses the damage causeddue to the stress factors after every ZC micro-cycle. This model is implemented through aprocess explained in the flowchart illustrated in Figure 4.6.

The SHS simulation starts as in the overall battery usage method, with SOH = 100%.The simulation is executed in 3 stages using the model. In Stage A, the battery activity isprobed within the SHS simulation. Zero-crossings (ZCs) are recorded, and the capacitydamage computations within this model start when the battery activity data is extractedfor a ZC-based micro-cycle. This includes evaluating the average battery DOD (DODi ),the energy throughput (Ethri ) and the average temperature (Ti ) for the i th ZCs micro-cycleas shown in Equations 4.7 and 4.8.

In Stage B, the total number of cycles n(Ti ,DODi ) possible at the DOD and tempera-ture levels is calculated using the polynomial lookup functions described in Section 4.3.1.Additionally, the proportional number of cycles (αi (Ethri ,Enom),DODi ) spent under thesame DOD and temperature levels for the i th ZCs micro-cycle is calculated as a function

4

82 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

StartPbattery pertime step

Battery active? ZC µ-cycle over? Evaluate DODi,Ti, Ethri

Evaluate: αi,n(αi), and battery

damage Di

SOH[i] =SOH[i − 1] - 0.2Di

SOH[i] < 80%?Extract time

elapsed (lifetime)Stop

yes yes

yes

nono

no

Stage A

Stage BStage C

Figure 4.6: Flowchart explaining the dynamic capacity fading model for lifetime estimation.

of the energy throughput (Ethri ) and the nominal battery capacity (Enom), as shown inEquation 4.10.

αi =Ethri

2×Enom ×DODi(4.10)

The damage (Di ) incurred to the battery life is then calculated using the Palmgren-Miner rule. This rule states that the lifetime of a component after undergoing a series ofload events is reduced by a finite fraction corresponding to each of the load events. Thisfraction is the ratio between the number of cycles the element has undergone under aparticular stress factor (or load event) to the total expected number of expected cyclesuntil EOL under that stress factor [9, 13]. Equation 4.11 represents the Miner’s rule.

D =E∑

i=1Di =

E∑i=1

αi

n(αi )(4.11)

Where αi : number of cycles spent under a stress factor σi (DOD and temperature), n(αi ):total number of cycles for the EOL to be reached, E : number of events taken place untilthe EOL condition is reached, D: total damage accumulated, and Di : damage at thebattery for each one of these events. This damage is then scaled and subtracted from thecurrent SOH. The damage scaling is done to ensure that when the total damage D = 1, theSOH is 80%.

Stage C involves checking for the EOL of the battery after every ZC micro-cycle ofactivity. If the SOH has reached 80% or below (or, in other words, if the damage D equalsa value of 1.), the simulation ends, and the SOH data is extracted as a function of time.If SOH is > 80%, the simulation continues in Stage A with the updated SOH. The stagesrepeat until the battery accumulates enough damage to reach its EOL. Note that the DODvalues are all dynamic, i.e., the DOD values are computed based on the updated batterycapacity after every micro-cycle.

SALIENT FEATURES OF THE DYNAMIC CAPACITY FADING MODEL

The dynamic capacity fading model is a significant upgrade over the overall batteryusage model both in complexity and accuracy with which the micro-degradation per ZCmicro-cycle is assessed. It has the following salient features.

4.4. RESULTS AND DISCUSSIONS

4

83

1. High resolution Given that the model captures the micro-cycle degradation, thedynamic capacity fading model serves with a high degree of resolution of the batterydegradation. Thus, the model can potentially capture the micro-degradation in atime-step as small as the time resolution of the input data used for the application.

2. Dynamic battery parameters Battery parameters like DOD and SOC are updatedafter every ZC micro-cycle based on the degraded battery capacity. Therefore, thebattery performance is not independent of the capacity fading.

3. SOH estimation The model allows for SOH to be estimated after every micro-cycle,which is not possible with the overall battery usage model.

4. Technology independent Given that the model is independent of electrochemicalprocesses for its implementation, it can be used for any battery technology. In thisstudy it is used to estimate the battery lifetime for 4 battery technologies.

5. Application independent Without any loss of generality, this model can be usedto estimate the battery lifetimes for other applications too, especially those appli-cations where the battery cycling undergoes cycling at similar C-rates as those ofSHS.

6. Usefulness at battery level Most performance-based experimentally constructedmodels are based on cell-level experiments, and the same battery technology mightexhibit different characteristics at the battery level. With the proposed model, onecan pick a candidate battery product and proceed to estimate the battery lifetimeat the system design stage.

4.4. RESULTS AND DISCUSSIONS

4.4.1. BATTERY USAGEThe battery experiences a wide range of DOD levels due to the intermittent nature of theincident solar irradiance on the SHS and the varying load profile. This can be seen fromFigure 4.7 that shows the normalized frequency of the different DOD levels throughoutthe year for the lead-acid gel battery under 2 cases, viz. considering the overall batteryDOD data and only the active battery DOD data.

The active battery data points were considered based on the zero-crossings, as de-scribed in Section 4.3.3. As seen in Figure 4.7, the two different cases differ in theirDOD levels. Between 10% to 70% DOD the active DOD occur relatively more frequently.However, in the very shallow (0–10%), and very deep DOD levels (70–80%), the trend isreversed, showing a lesser occurrence of active DODs, underlining the periods of inactivityof the battery. This corresponds to when the battery was either full (not needed to powerthe load) or empty (could not power the load). Moreover, the maximum limit of the DODshown in the histogram is 80%, owing to the lower limit of the battery SOC fixed at 20% inthe simulation.

Additionally, the battery cycling related data is extracted using both the coarse averageand the ZCs approach. The results are shown in Table 4.2.

4

84 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

0–1010–20

20–3030–40

40–5050–60

60–7070–80

0.05

0.1

0.15

0.2

0.25

DOD [%]

No

rmal

ized

freq

uen

cy[-

] All Active

Figure 4.7: Normalized frequency of the DODs experienced by the lead-acid gel battery for both the overallbattery DOD data and only the active battery DOD data.

A difference can be seen across the 4 battery technologies for the various batteryparameters. In general, the coarse average-based DOD is more optimistic than the ZCs-based DOD for all the technologies. The difference in the DOD using the same method fordifferent batteries comes from the fact that the different batteries have different operatingefficiency, and therefore need to cycle differently in order to meet the storage demandof the given SHS application. Similarly, the energy throughputs of the different batterytechnologies are also different. The coarse-average temperature is the same for everybattery technology based on the meteorological data, while the ZCs-based temperature isslightly different, depending on the duration of the ZC-cycles (Equation 4.8), which differacross technologies.

LiFePO4 fares the best across the different battery statistics, while the NiCd batteryfares the worst. Lead-acid gel battery performs slightly better than its flooded counterpart.In general, the more efficient the battery, the lower the DOD as well as temperature andenergy throughput for the exact same application. Consequently, the cycle life will alsobe higher than that of a battery with lower efficiency, as seen in Section 4.4.2 below.

Even though the average DOD values between the two methods differ only by about1–1.5%, this difference can be significantly amplified based on the battery size chosen andthe SHS load requirements. For example, oversizing the battery (lower LLP) would haveresulted in a larger difference between the coarse average and the ZCs-based active DODcalculations. Going forward, the more conservative ZCs-based battery lifetime resultsare discussed, as using a coarse average-based estimation can yield to an optimisticprediction of battery lifetime.

4.4. RESULTS AND DISCUSSIONS

4

85

Table 4.2: Usage statistics for the 4 battery technologies over 1 year of SHS simulation. LA: lead-acid, CA: Coarseaverage-based, ZC: Zero crossings-based.

Battery DOD (%) T (C) Ethr

CA ZC CA ZC (kWh/year)

Flooded LA 37.81 38.21 27.16 26.94 613.9LA gel 36.23 36.73 27.16 26.78 589.7NiCd 39.77 40.04 27.16 27.2 647.7LiFePO4 35.11 35.66 27.16 26.7 575.7

4.4.2. LIFETIME ESTIMATIONBased on the battery usage discussed in Section 4.4.1, the battery lifetime is estimatedbased on the methodologies described in Section 4.3.

OVERALL BATTERY USAGE-BASED LIFETIME ESTIMATION

The results for the overall battery usage-based estimated lifetimes are shown in Table 4.3.

Table 4.3: Cycle life and battery lifetime in years, based on the overall battery usage-based lifetime estimation.LA: Lead-acid.

Technology Cycle life (-) Lifetime (years)

Flooded LA 3329 6LA gel 3796 6.8NiCd 1662 3LiFePO4 16450 29.4

Following the battery usage statistics discussed in Section 4.4.1, the battery lifetimefollows a similar relative trend. LiFePO4 comprehensively outperforms the other batterytechnologies, having an estimated battery lifetime of nearly 5 times the lead-acid batteriesand almost 10 times the NiCd battery.

DYNAMIC LIFETIME ESTIMATION

Based on the dynamic lifetime estimation method described in Section 4.3.4, the batterylifetimes for the SHS use-case were estimated for the 4 battery technologies. The resultsfor the dynamic capacity fading-based lifetime estimation are shown in Figure 4.8. Forcomparison, the results from the overall battery usage-based estimation method are alsoplotted.

As seen in Figure 4.8, the dynamic lifetime estimation is found to be much moreconservative, with lifetimes for LA-gel, LA-flooded, NiCd, LiFePO4 batteries as 5.6, 5.1,2.86, and 16.7 years, which are 18%, 14%, 3%, and 43% less than the corresponding overallusage-based lifetime estimates, respectively. This is attributed to the fact that dynamiclifetime degradation captures the micro-degradation of the battery capacity for everymicro-cycle. Consequently, the battery is progressively degraded as it enters the next

4

86 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

LA-gel LA-flooded NiCd LiFePO4

5

10

15

20

25

30

5.6 5.1

2.86

16.7

Battery Technology [-]

Bat

tery

life

tim

e[y

ears

]

Overall Dynamic

Figure 4.8: Dynamic battery lifetime estimation results for the 4 battery technologies in comparison with theresults using the overall battery usage method.

micro-cycle with a faded capacity, instead of calculating the lifetime based on overallusage of one year where the battery performs identically without any degradation. There-fore, the dynamic capacity fading model-based lifetime estimation is considered morepractical and realistic. Additionally, the longer the battery lasts, the more pronouncedeffect of dynamic capacity fading. Therefore, from NiCd to LiFePO4 the deviation fromthe overall battery lifetime estimation method is increasing.

Of course, it must be noted that the methodology described here relies on the manu-facturer’s datasheet, and therefore the relative lifetime estimation for this particular SHSapplication need not be extendable to all battery products from the same technology.This is because different battery products from the same technology often display varyingcycle lives depending on the construction geometry, propriety manufacturing processes,amongst others.

A comparison of this model is also made with an experimentally obtained empiricalmodel for capacity fading of LiFePO4, as explained in Section 4.4.3.

4.4.3. COMPARISON WITH AN EMPIRICAL BATTERY LIFETIME ESTIMATION

MODEL

As an experimental validation of the dynamic lifetime estimation model at the batterypack level can take an impractically long time, the accuracy of the model was comparedto the results obtained from another experimentally created, empirical-based lifetimeestimation model based on LiFePO4 battery technology as described in [39]. The empiricalmodel can been described by the Equations 4.12 and 4.13 [39].

4.4. RESULTS AND DISCUSSIONS

4

87

c f =E∑i

((ks1SOCdev,i .eks2.SOCav g ,i +ks3.eks4.SOCdev,i )

.e

(− Eg

R

(1

Ti− 1

Tr e f

)))Ahi (4.12)

Where c f is the capacity fading experienced by the battery due to all the cycling events E,i is an event for an arbitrary length of time, Ahi is the total charge processed during eventi , E is the total number of events.The temperature dependence term comes from the Arrhenius equation, where ks1 to ks4

are constants determined experimentally and reported in [39], SOCdev,i is the normalizedstandard deviation from SOCav g in event i . This value is derived from Equation 4.13 [5].

SOCdev =

√√√√√√ 3

∆Ahm

Ahm∫Ahm−1

(SOC (Ah)−SOCav g )2.d Ah (4.13)

To be applied in the SHS application discussed in this study, Equation 4.12 wasadapted to the battery usage obtained in the SHS use-case. Similar to the approachdiscussed in Section 4.3.4, the capacity fading was calculated per cycling event (ZCmicro-cycle) with the help of Equations 4.12 and 4.13. Consequently, the battery lifetimewas found to be 14.3 years, about 14.2% deviation from the dynamic battery lifetimeestimation method.

The State of Health (SOH) is also calculated based on the empirical method, which isdiscussed along with the other SOH results in Section 4.4.3 below.

STATE OF HEALTH (SOH)While the lifetime results mentioned above give a measure of the overall battery lifefor the given application, the SOH results over time give a measure of the rate of thecapacity fading for different battery technologies. For the given battery lifetime estimationmodelled in this study, the State of Health (SOH) was computed for the different batteriesunder operating conditions, i.e., taking only cyclic ageing into account.

The results are plotted in Figure 4.9. For the same starting battery capacity, the slowestfading rate is seen to be experienced by the LiFePO4 battery, while the fastest rate (steepestslope) is experienced by the considered NiCd battery. The SOH results seem to show anear-linear trend over time.

Additionally, the SOH is plotted for the empirical lifetime estimation model for thesame SHS use-case considered, as shown in Figure 4.9. The SOH from the empiricalmodel follows the dynamic lifetime estimation method-based SOH (both for LiFePO4)closely, especially up to 89% SOH. After year 10, the SOH estimates start to deviate. TheSOH deviations at the various year-marks differ by a maximum of 2.8%. However, it mustbe noted that the comparison with empirical model is only valid for the LiFePO4 batterytechnology and as such the degree of accuracy shown by this comparison cannot beextended to other technologies solely on this basis. No empirical models were found forthe other technologies at the time of writing this chapter, which could enable a similar

4

88 4. ESTIMATING BATTERY LIFETIME IN SOLAR HOME SYSTEMS

0 5 10 15 20

0.8

0.85

0.9

0.95

1

EOL

Time [years]

SOH

[-]

NiCDLA-floodedLA-gelLiFePO4

Empirical (LiFePO4)

Figure 4.9: SOH for the different battery technologies as estimated by the dynamic lifetime method. The blackdashed line denotes the end of life.

comparison for the SHS use-case for the exact same stress factors as considered in theproposed methodology.

In summary, the main difference between the 2 proposed models, viz. the overallusage model, and the dynamic capacity fading model, is the increased complexity andaccuracy of the dynamic capacity fading model. Within the overall battery usage model,the coarse average approach is rather crude but quick, while the ZC micro-cycle-basedapproach is more computationally demanding. However, the overall battery usage modelis not without its assumptions, as discussed in Section 4.3.3. The dynamic capacity fadingmodel is the most complex but remarkably matches the experimentally derived empiricalmodel until about 89% SOH.

4.4.4. RELEVANCE FOR SHS DESIGN

In the context of SHS application, the battery lifetime estimation exercise is quite relevant,as mentioned previously in Section 4.1. This is even more relevant in the off-grid electrifi-cation segment, where system reliability plays a critical role in technology adoption.

Compared to a typical PV module that lasts 25 years, an SHS battery usually lastsmuch less. Therefore, the battery will need to be replaced in the SHS lifetime. Havinga larger battery size leads to lower levels of average DOD while cycling. Thus, a largerbattery size not only increases the lifetime but also consequently reduces the number ofbattery replacements needed in the SHS lifetime. This presents an interesting trade-offbetween the upfront costs and battery replacement costs; the cost of having an increasedlifetime (and therefore increased battery size) for the same application is the increasedupfront costs that come along with a larger battery size. Therefore, estimating the batterylifetime at the SHS design stage can help optimize this delicate trade-off between upfrontcosts and battery replacement costs. The methodology described in this chapter was

4.5. CONCLUSION

4

89

used to investigate the impact of SHS battery sizing on battery lifetime to underline theimportance of estimating battery lifetime at the SHS design stage. This can be found inChapter 5, where SHS is optimally sized while taking battery lifetime into account.

4.5. CONCLUSIONThis chapter described a practical methodology for estimating the battery lifetime withoutneeding to model the electrochemical processes within the battery or needing dedicatedexperiments. Moreover, this methodology uses available battery data from the man-ufacturer for candidate battery technologies at hand. The methodology is applicableirrespective of the battery technology, as it is independent of the technology-specificelectro-chemical processes. The methodology can also be extended to other applicationsexperiencing similar battery C-rates without loss of generality. The described method-ology is expected to help SHS designers make informed decisions with respect to thebattery storage at the system sizing stage.

Dynamic capacity fading model introduced in this chapter was deemed more practicalthan the overall battery usage model. For a given load profile and specific SHS system size,the estimated lifetimes for LA-gel, LA-flooded, NiCd, LiFePO4 batteries were 5.6, 5.1, 2.86,and 16.7 years, respectively using the dynamic capacity fading model. Comparison of thisproposed dynamic model with an experimentally derived empirical model of LiFePO4

battery yielded very close results, with the SOH values over time being within 2.8% ofeach other.

The battery lifetime estimation methodology described in this chapter is also usedin Chapter 5 to perform a multi-objective optimization for determining optimal systemsizes for SHS.

REFERENCES[1] Good Solar Initiative, Quality Charter: Quality Standards for Accredited Solar Suppli-

ers. Good Solar Initiative, 2016.

[2] A. A. Phadke, A. Jacobson, W. Y. Park, G. R. Lee, P. Alstone, and A. Khare, “Powering ahome with just 25 watts of solar pv. super-efficient appliances can enable expandedoff-grid energy service using small solar power systems,” tech. rep., Lawrence Berke-ley National Laboratory (LBNL), Berkeley, CA (United States), 2015.

[3] N. Narayan, J. Popovic, J. C. Diehl, S. Silvester, P. Bauer, and M. Zeman, “Developingfor developing nations: Exploring an affordable solar home system design,” in 2016IEEE Global Humanitarian Technology Conference (GHTC), pp. 474–480, Oct 2016.

[4] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrifica-tion,” in 2017 International Conference on the Domestic Use of Energy (DUE), pp. 8–15,April 2017.

4

90 REFERENCES

[5] N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P. Bauer, and M. Zeman,“A simple methodology for estimating battery lifetimes in solar home system design,”in 2017 IEEE AFRICON, pp. 1195–1201, Sept 2017.

[6] C. Zou, C. Manzie, D. Nešic, and A. G. Kallapur, “Multi-time-scale observer designfor state-of-charge and state-of-health of a lithium-ion battery,” Journal of PowerSources, vol. 335, pp. 121–130, 2016.

[7] C. Zou, X. Hu, Z. Wei, and X. Tang, “Electrothermal dynamics-conscious lithium-ion battery cell-level charging management via state-monitored predictive control,”Energy, vol. 141, pp. 250 – 259, 2017.

[8] C. Zou, X. Hu, Z. Wei, T. Wik, and B. Egardt, “Electrochemical estimation and controlfor lithium-ion battery health-aware fast charging,” IEEE Transactions on IndustrialElectronics, vol. 65, pp. 6635–6645, Aug 2018.

[9] V. Marano, S. Onori, Y. Guezennec, G. Rizzoni, and N. Madella, “Lithium-ion batterieslife estimation for plug-in hybrid electric vehicles,” in 2009 IEEE Vehicle Power andPropulsion Conference, pp. 536–543, Sept 2009.

[10] M. Dubarry, B. Y. Liaw, M.-S. Chen, S.-S. Chyan, K.-C. Han, W.-T. Sie, and S.-H. Wu,“Identifying battery aging mechanisms in large format li ion cells,” Journal of PowerSources, vol. 196, no. 7, pp. 3420 – 3425, 2011.

[11] V. Agarwal, K. Uthaichana, R. A. DeCarlo, and L. H. Tsoukalas, “Development andvalidation of a battery model useful for discharging and charging power control andlifetime estimation,” IEEE Transactions on Energy Conversion, vol. 25, pp. 821–835,Sept 2010.

[12] R. Dufo-López, J. M. Lujano-Rojas, and J. L. Bernal-Agustín, “Comparison of differentlead-acid battery lifetime prediction models for use in simulation of stand-alonephotovoltaic systems,” Applied Energy, vol. 115, pp. 242 – 253, 2014.

[13] M. A. Tankari, M. B. Camara, B. Dakyo, and G. Lefebvre, “Use of ultracapacitorsand batteries for efficient energy management in wind -diesel hybrid system,” IEEETransactions on Sustainable Energy, vol. 4, pp. 414–424, April 2013.

[14] M. Musallam and C. M. Johnson, “An efficient implementation of the rainflow count-ing algorithm for life consumption estimation,” IEEE Transactions on Reliability,vol. 61, pp. 978–986, Dec 2012.

[15] T. Dragicevic, H. Pandzic, D. Skrlec, I. Kuzle, J. M. Guerrero, and D. S. Kirschen,“Capacity optimization of renewable energy sources and battery storage in an au-tonomous telecommunication facility,” IEEE Transactions on Sustainable Energy,vol. 5, pp. 1367–1378, Oct 2014.

[16] T. M. Layadi, G. Champenois, M. Mostefai, and D. Abbes, “Lifetime estimation tool oflead-acid batteries for hybrid power sources design,” Simulation Modelling Practiceand Theory, vol. 54, pp. 36 – 48, 2015.

REFERENCES

4

91

[17] D. G. für Sonnenenergie, Planning and Installing Photovoltaic Systems: A Guide forInstallers, Architects and Engineers. Planning and Installing, Taylor & Francis, 2008.

[18] P. Mohanty, T. Muneer, and K. Mohan, Solar Photovoltaic System Applications.Springer, 2016.

[19] M. R. Palacin and A. de Guibert, “Why do batteries fail?,” Science, vol. 351, no. 6273,pp. 1253292–1253292, 2016.

[20] P. Keil, S. F. Schuster, J. Wilhelm, J. Travi, A. Hauser, R. C. Karl, and A. Jossen, “CalendarAging of Lithium-Ion Batteries,” Journal of The Electrochemical Society, vol. 163, no. 9,pp. A1872–A1880, 2016.

[21] M. T. Lawder, P. W. C. Northrop, and V. R. Subramanian, “Model-Based SEI LayerGrowth and Capacity Fade Analysis for EV and PHEV Batteries and Drive Cycles,”Journal of The Electrochemical Society, vol. 161, no. 14, pp. A2099–A2108, 2014.

[22] P. Ruetschi, “Aging mechanisms and service life of lead-acid batteries,” Journal ofPower Sources, vol. 127, no. 1-2, pp. 33–44, 2004.

[23] K. Brik and F. Ben Ammar, “Causal tree analysis of depth degradation of the lead acidbattery,” Journal of Power Sources, vol. 228, pp. 39–46, 2013.

[24] J. C. Viera, M. González, J. C. Anton, J. C. Campo, F. J. Ferrero, and M. Valledor, “NiMHvs NiCd batteries under high charging rates,” INTELEC, International Telecommuni-cations Energy Conference (Proceedings), 2006.

[25] J. Wang, P. Liu, J. Hicks-Garner, E. Sherman, S. Soukiazian, M. Verbrugge, H. Tataria,J. Musser, and P. Finamore, “Cycle-life model for graphite-LiFePO4 cells,” Journal ofPower Sources, vol. 196, no. 8, pp. 3942–3948, 2011.

[26] Hoppecke, Germany, Installation , commissioning and operating instructions forvalve-regulated stationary lead-acid batteries - Solar Battery Data Sheet, 2013.

[27] Valence | Lithium Ion Phosphate Battery Manufacturer, “Valance battery cyclelifecurves.” url=https://www.valence.com/why-valence/long-lifecycle/, 2017. Date lastaccessed: 2017-05-03.

[28] Hoppecke, Germany, Installation , commissioning and operating instructions forvented stationary lead-acid batteries - Solar Battery Data Sheet, 2013.

[29] Alpha Technologies, Inc., United States, Alpha Power - Industrial Ni-Cd BatteriesStandard Range Technical Manual, 2009.

[30] M. Bhatia and A. Nicolina, Capturing the Multi-Dimensionality of Energy Access. LiveWire, 2014/16. World Bank, Washington, DC. ©World Bank, 2014.

[31] Global LEAP, “The State of the Off-Grid Appliance Market,” Tech. Rep., 2016.

4

92 REFERENCES

[32] H. Olk and J. Mundt, “Photovoltaics for productive use applications. a catalogueof dc-appliances,” Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ)GmbH, 2016.

[33] Meteotest, “(software) meteonorm ver 7.1,” 2014.

[34] B. Dunn, H. Kamath, and J.-m. Tarascon, “Electrical energy storage for the Grid : ABattery of choices,” Science Magazine, vol. 334, no. 6058, pp. 928–936, 2011.

[35] P. Van den Bossche, F. Vergels, J. Van Mierlo, J. Matheys, and W. Van Autenboer,“SUBAT: An assessment of sustainable battery technology,” Journal of Power Sources,vol. 162, no. 2 SPEC. ISS., pp. 913–919, 2006.

[36] K. Hojckova, J. Jelinek, M. Schneider, N. Spittler, and I. Varju, “Evaluation of batterystorage technologies for sustainable and rural electrification in sub-saharan africa,”Regional Academy on the United Nations, 2014.

[37] X. Luo, J. Wang, M. Dooner, and J. Clarke, “Overview of current development inelectrical energy storage technologies and the application potential in power systemoperation,” Applied Energy, vol. 137, pp. 511–536, 2015.

[38] R. M. Herritty and J. Midolo, “Nickel cadmium batteries for photovoltaic appli-cations,” in Thirteenth Annual Battery Conference on Applications and Advances.Proceedings of the Conference, pp. 255–257, Jan 1998.

[39] L. Lam, “A practical circuitbased model for state of health estimation of liion batterycells in electric vehicles,” Master’s thesis, Delft University of Technology, August2011.

5EXPLORING SHS BOUNDARIES FOR

ELECTRIFICATION: OPTIMAL SHSSIZING

As long as poverty, injustice and gross inequality persist in our world,none of us can truly rest

Nelson Mandela

ABSTRACT

Optimal system sizing for SHS is a vital task as both oversizing and undersizing a systemcan be detrimental to system cost and power availability, respectively. This chapterpresents an optimal SHS sizing methodology that minimizes the loss of load probability(LLP), excess energy dump, and battery size while maximizing the battery lifetime. Agenetic algorithm-based multi-objective optimization approach is utilized to determinethe optimal SHS sizes. The potential for SHS to cater to every tier of the Multi-tierframework (MTF) for measuring household electricity access is examined. The optimalsystem sizes for standalone SHS are found for different LLP thresholds. Results show thatbeyond tier 2, the present day SHS sizing needs to be expanded significantly to meet theload demand. Additionally, it is deemed untenable to meet the electricity needs of thehigher tiers of MTF purely through standalone SHS without compromising one or moreof the system metrics. A way forward is proposed to take the SHS concept all the way upthe energy ladder such that load demand can also be satisfied at tier 4 and 5 levels.

This chapter is based on: N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, & M.Zeman. (2019). Exploring the boundaries of Solar Home Systems (SHS) for off-grid electrification: Optimal SHSsizing for the multi-tier framework for household electricity access. Applied Energy, 240, 907-917.

93

5

94 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

OUTLINE

This chapter is divided into five sections. Section 5.1 introduces the chapter, underliningthe motivation for optimal system sizing and presenting a literature review. Section 5.2describes the various parameters and metrics used in the study. Section 5.3 details all thesteps considered in the methodology, while Section 5.4 discusses the results and theirimplication on the utility of SHS as a solution for electrifying higher tiers of the MTF.Finally, Section 5.5 concludes the chapter with the closing thoughts.

5.1. INTRODUCTIONIn this chapter, an optimal sizing methodology is introduced to optimally size SHS forevery tier of the MTF, which gives insights on the level of PV and battery storage neededto enable electrification across the various tiers.

5.1.1. IMPORTANCE OF OPTIMAL SHS SIZING

SHS sizing comprises the PV sizing, battery sizing, and the sizing of the power converters.PV sizing mainly depends on the total energy needed from the PV generator, which inturn depends on the load profile. A lower than adequate PV size results in system failureor a high number of loss of load events, i.e., high loss of load probability (LLP). LLP isa system metric that quantifies the system’s reliability in meeting the load demand, asexplained in detail in Section 5.2.1. A higher than adequate PV size, however, increasesthe amount of energy dumped — or the non-utilization of solar energy — when the loadis satisfied and the battery is full.

Power converter sizing mainly depends on the peak power of the PV and the load.Some degree of dimensioning flexibility can be achieved depending on the undersizing ofthe converter. A lower than peak power of the converter may not be suited for peak poweroperation but can perform more efficiently at most of the other operating points. Thisdepends mainly on the frequency of occurrence of the different power levels expectedthroughout the operation of the converter. Thus, a suitably lower than peak sized con-verter can lead to cost savings. Compared to PV and power converter, however, batterysizing is far more involved.

The battery is a vital component of the SHS that not only enables energy storage ofthe PV output but also caters to the load when there is no solar generation. However,the battery is the most expensive SHS component while suffering from low lifetimes ascompared to other SHS components. Additionally, a smaller than adequate battery sizewill result in failure to meet the load requirements (high LLP), while an oversized batterywill drastically increase the upfront costs of the system. Also, a larger battery size canlead to lower depth of discharge (DOD) levels, and therefore higher lifetime, whereas asmaller battery size can lead to lower lifetimes due to the deep DOD levels [1, 2]. Usualbattery lifetimes are much lower than typical PV module lifetime of 25 years. Therefore, ahigher battery lifetime is advantageous as it means fewer replacements during the SHSlifetime. Battery costs and lifetime thus have an intricate relationship, making batterysizing extremely relevant albeit challenging in SHS design. Battery sizing and lifetime can,therefore, be considered as critical parameters when dimensioning an SHS.

An optimal SHS size can thus be considered as one that results from an SHS dimen-

5.1. INTRODUCTION

5

95

sioning exercise that minimizes the LLP, excess energy, and battery size while maximizingthe battery lifetime.

5.1.2. LITERATURE STUDY

In general terms, most sizing-based studies utilize sizing criteria that result in a goodtrade-off between the system reliability (or power availability) and system investment cost.Previous studies have dealt with the sizing problem by only looking at one or two metrics,including LLP [3, 4], system cost, and battery lifetime [5]. LLP has been considered as anobjective in [6] as well as a constraint. In [7], the main objective function was to minimizethe total SHS investment cost, while keeping LLP ≤ 2%. LLP was also used as the primarysizing criterion for an off-grid PV-battery system in Bolivia [8], where the system size wasdetermined for three different case studies: a household, school, and health center.

Additionally, battery lifetime has been included as a key parameter in a single weightedobjective optimization for a PV array, diesel generator and battery system to minimizebattery degradation and fuel consumption [9]. Similarly, the optimal sizing of a PV-battery-diesel generator hybrid system has been investigated [10], paying special attention tosystem cost and environmental impact of the off-grid system. In this case, the levelizedcost of electricity and the carbon footprint of energy are defined as the main metrics thatmust be minimized.

Other sizing methods are based on intuitive “rules of thumb". The concept of daysof autonomy (DOA) for battery sizing, which consists of finding the storage size thatcan fulfill the load for a predefined amount of days in the absence of solar generationis an example [11]. Another example is the concept of nights of autonomy (NOA). Boththe DOA and NOA concepts are discussed in Section 5.2.2. More complex optimizationmethods have been introduced over the past years, especially iterative optimizationapproaches where the system performance for the objective is iteratively evaluated acrossthe decision variable space [12, 13]. For instance, a standalone PV-battery system hasbeen dimensioned based on loss of power probability and life cycle cost using an iterativeprocedure in [14].

However, the main drawback of these techniques is that the system is only optimizedbased on one objective function. Therefore, a system with multiple trade-offs, or objectivefunctions, has not been completely tackled, especially for the application of solar homesystems. In comparison, multi-objective optimization (MOO) techniques offer betterapplicability. Nowadays, MOO techniques based on artificial intelligence are widely used,which are generally more robust, and are better equipped to deal with multi-objectiveoptimization problems [15]. Among them, Genetic Algorithms (GA) are powerful meta-heuristic techniques that are capable of reaching global optima with high accuracy andappropriate computational speed [16]. Former studies using MOO techniques in PVsystems show different focus areas compared to our study. In [17], authors quantify thetrade-offs between economic and environmental performances of rural solar PV projects.Authors in [18] use a sizing algorithm based on Particle Swarm Optimization (PSO) fora PV-battery-hydrogen fuel cell-based hybrid system. There is a clear research gap inliterature with respect to studies focusing on optimal sizing of present-day SHS, especiallytaking into account battery lifetime as one of the objectives.

In this study, the different SHS parameters need to be optimized simultaneously,

5

96 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

making the MOO a necessity. Consequently, in this study, we use a GA algorithm for thetask of optimal sizing of an SHS for different electrification levels defined by the MTF.In this optimization, the PV rating, battery capacity, and converter sizes are obtainedtaking into account the most optimum combination of LLP, excess energy, battery size,and battery lifetime.

5.1.3. CONTRIBUTIONS OF THIS CHAPTER

Following are the main contributions of this chapter.

1. An optimal sizing methodology for SHS is presented that minimizes the LLP, excessenergy, and battery size while maximizing the battery lifetime specific to SHSapplications.

2. For the first time, optimal SHS sizing is investigated for the various tiers of MTF forhousehold electricity access.

3. Inadequacies of standalone SHS are highlighted for higher tiers of electrificationand a possible alternative is proposed for climbing the so-called electrificationladder.

5.2. SYSTEM METRICS AND PARAMETERS

5.2.1. SYSTEM METRICS

The system metrics used in this work are discussed below.

LOSS OF LOAD PROBABILITY

The Loss of Load probability is a measure of the system downtime. It is defined as the ratioof the amount of time the system fails to deliver the demanded power to the total amountof operation time the system was designed to deliver power for [4]. For modelling andsimulating off-grid systems, this is an application specific or a user-defined constraint.For example, in [4], the LLP over a year was constrained to 1.8%, and in [19], a similarmetric had a minimum limit of 2%.

LLP =∑N

i=1 tdowntime,i

N(5.1)

where tdowntime,i takes a value of 1 if the system fails to deliver the expected loaddemand in the i th time interval, and 0 if the system fully meets the load requirement; Nis the time period of interest. Since the system is modelled with a 1-min data resolution,tdowntime,i is updated every minute, while a 1-year long period of interest yields an N valueof 525600.

The choice of LLP can also be dependent on the application. For instance, [11]states that the recommended LLP values for domestic illumination, appliances, andtelecommunications applications are 0.01, 0.1, and 0.0001 respectively.

5.2. SYSTEM METRICS AND PARAMETERS

5

97

UNSATISFIED LOAD ENERGY (Efail)Efail quantifies the unmet energy demand in kWh over a given period of time. In thisstudy, it is mathematically defined as the summation of the unsatisfied energy over a yearfor each time interval i , as shown in Equation 5.2.

Efail =N∑

i=1Eunsatisfied,i (5.2)

ENERGY DUMP (Edump) AND DUMP RATIO (Rdump)This is the total amount of energy that is unused when the local battery is full and theload demand is met while the PV can still generate more power. In this study, this value isconsidered over a period of 1 year for the system. In order to make relative comparisonseasier, a term dump ratio is introduced, which is the ratio between the total energy dumpof the system in a year divided by the annual load energy need, as seen in Equation 5.3.

Rdump = Edump

Eload,year(5.3)

SHS SIZE

SHS size corresponds to the electrical dimensions of PV, battery and the power converters.Typically, this is specific to the rated power of the PV in Wp, total battery capacity in Wh,and peak power rating in W of the converter.

BATTERY LIFETIME

This is the lifetime in years of operation after which the battery capacity reduces to below80% of its nominal rated capacity. At the end of this period, the battery is said to havereached its end of life (EOL).

5.2.2. SYSTEM PARAMETERSThe following are the other important system parameters that are referred to in thischapter.

STATE OF CHARGE AND DEPTH OF DISCHARGE

This has already been described in Chapter 4 in Section 4.2.1.

STATE OF HEALTH

State of Health (SOH) is indicative of the fraction of the nominal rated battery capacityactually available for cycling.

DAYS OF AUTONOMY (DOA)Days of autonomy refers to the number of days a particular battery size is capable ofpowering the full load demand assuming there is no PV generation. This is often used inapproximate battery sizing. For example, a battery size in Wh can potentially be writtenas shown in Equation 5.4.

Ebatt =Eload ·nDOA

DOD ·ηbatt

(5.4)

5

98 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

where Eload is the load energy requirement over a day, nDOA is the number of days ofautonomy, ηbatt is the average battery efficiency, and DOD is the average DOD of theintended battery cycling. Practitioners often use DOA or NOA to size the battery as a “ruleof thumb”. However, this can lead to inaccurate and suboptimal system sizing, as shownlater in Section 5.4.1.

NIGHTS OF AUTONOMY (NOA)Nights of autonomy refers to the number of nights (night = non-sunlight hours of theday) where a battery can completely power the load demand assuming no PV generationto recharge the battery. This is also used in approximate battery sizing. For example, abattery size in Wh can potentially be written as shown in Equation 5.5.

Ebatt =Enightload ·nNOA

DOD ·ηbatt

(5.5)

where Enightload is the load energy requirement in the non-sunlight hours in a day, andnNOA is the number of nights of autonomy.

5.3. METHODOLOGYThe various steps employed in the methodology of this study are described in this section.Section 5.3.1 presents the inputs to the SHS model, Section 5.3.2 presents the archi-tecture of the system, Section 5.3.3 describes a dynamic PV output modeling method,Section 5.3.4 describes the steps involved in battery lifetime modelling, Section 5.3.5discusses the power management scheme used in the standalone SHS, Section 5.3.6 de-scribes the steps used to evaluate the converter rating, and Section 5.3.7 details a geneticalgorithm (GA)-based multi-objective optimization (MOO) approach for SHS sizing.

5.3.1. INPUTS TO THE SHS MODEL

LOCATION AND METEOROLOGICAL INPUTS TO THE SHS MODEL

Since the ground level meteorological data was not readily available for the remote ruralareas, the data for an Indian city (with 5 Equivalent sun hours of irradiance per day) wastaken from the Meteonorm software database [20]. The meteorological data is assumedto be representative of other remote areas in similar latitudes in the tropical regions. Thedata used in this study has a resolution of 1 minute.

The main meteorological inputs used in the SHS model are: (a) Ambient temperature(Tamb), (b) Direct Normal Irradiance (GDNI), (c) Diffused Horizontal Irradiance (GDHI),(d) Global Horizontal Irradiance (GGHI), and (e) Wind speed (vw).

LOAD PROFILES FOR THE MTFStochastic load profiles constructed for the various tiers of the MTF in Chapter 3 wereused as inputs in this study. Figure 5.1 shows a sample load profile for a representativeday from the constructed 1-year load profile for a tier 2 household. The household-levelload datasets for all the tiers have been obtained from [21]. Table 3.4 summarizes theimportant load profile characteristics for each tier of the MTF.

5.3. METHODOLOGY

5

99

0 4 8 12 16 20 240

10

20

30

40

50

Time [h]

Pow

er[W

]

Figure 5.1: 1-day load profile of an off-grid household with tier 2 electricity access. The total energy consumptionfor this day is 257 Wh with a peak of around 49 W. Data sourced from [21].

5.3.2. MODULAR SHS ARCHITECTUREClimbing up the so-called rural electrification ladder requires expansion of the off-gridsystem to cater to increased load demand. A modular architecture for SHS is thereforeproposed that could potentially allow for expansion of the system at the household level.Figure 5.2 shows the modular architecture.

PV

BatteryLoads

DC

Battery

bus

DC

Figure 5.2: Modular DC architecture for an SHS that enables intra-household growth option. Dashed batterywith converter illustrates the modular capability.

All the SHS components, viz., PV, battery, and DC loads, are connected to the central

5

100 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

DC bus via converters. Expansion of the system can be achieved by adding more of theSHS components to the DC bus via a converter, as shown in Figure 5.2 with the dashedbattery at the bottom. The DC bus can be operated at 12 V, 24 V or 48 V. As more highpower loads are connected, a higher bus voltage is recommended to keep the current andtherefore cable losses low.

5.3.3. DYNAMIC PV OUTPUTThe dynamic PV output for each minute is calculated based on an elaborate PV model asillustrated in Figure 5.3.

Orientationoptimization

sun positionalgorithm

Module temper-ature estimation

Efficiencycorrection

GPOA

Tmod

Es, As

GDNI, GDHI, GGHI

Tamb, vw

ηSTC, APV, k

Location coordinates

ηdyn, PPV(t), EPV (PV yield)Inputs

(i)

(ii)

(iii)

(iv)

Figure 5.3: Flowchart explaining the steps involved in calculating the dynamic PV output.

There are 4 different kinds of inputs used. Firstly, the location coordinates in terms oflatitude and longitude are used to determine the sun position throughout the year. Thetwo main parameters that quantify the sun position are the sun azimuth (Es) and the sunaltitude (As).

Secondly, the ground level irradiance data for the location is used, as mentioned inSection 5.3.1. Along with Es and As, the PV module orientation optimization model usesGDNI, GDHI, and GGHI to evaluate the module azimuth and tilt that maximizes the planeof array irradiance (GPOA). Additionally, the GPOA is computed for the optimal azimuthand tilt.

The dynamic efficiency (ηdyn) of the PV module can be quite different than its ratedefficiency (ηSTC). As ηdyn depends on both the module temperature (Tmod) and GPOA, themodule temperature needs to be estimated. The fluid dynamic model is used to estimateTmod by using Tamb, wind speed (vw), and the calculated GPOA as inputs [11] .

Finally, the PV efficiency is corrected and the dynamic efficiency ηdyn is calculatedusing the module area (APV), the temperature coefficient of power for the PV module (k),and ηSTC. The selected PV module is Jinko Solar JKM265P. Although the rated power is 265

More details can be found in Appendix B

5.3. METHODOLOGY

5

101

Start

Pexcess = PPV −Pload − Ploss

Pexcess > 0SOC = 100%

Dump sur-plus energy

Charge the batterywith surplus

End End

Deficit < deliverable Ebatt

Feed loadwith battery

End

Deliverable Ebatt > 0

Feed load partially

End End

yes

Load fully met

yes

no

Surplus=0

Load fully met

yes

Load fully met

no no

yes

no

unmet load

Figure 5.4: Flow chart explaining the power management scheme used in the standalone SHS model; it rep-resents the algorithm followed in every time step. Note that dumped energy could be viewed as PV powerproduction curtailment in practice, leading to potential under-utilization of solar energy.

Wp, a normalized dynamic PV output scaled down to 10 Wp is considered for the optimalsizing methodology. The detailed methodology for modelling the dynamic PV output hasbeen described in a previous work done by the authors in [4].

5.3.4. ESTIMATING BATTERY LIFETIME

In this study, a valve regulated lead acid (VRLA) battery is considered as the energy storagedevice. The practical battery lifetime estimation method presented in Chapter 4 is used inthis study. As explained in Chapter 4, the model uses battery manufacturers’ datasheetsto refer to the cycle life curves at different temperatures [22]. A constant round-tripefficiency of 90% is considered for the VRLA battery [1].

For each micro-cycle of battery activity determined by a current zero-crossing, theproportional number of cycles spent under the DOD and temperature levels, is thenevaluated, as shown in Equation 4.10. The partial damage incurred by the battery in eachmicro-cycle is calculated as shown in Equation 4.11. This damage is then scaled andsubtracted from the present SOH, so that when the cumulative damage becomes 1, theSOH reaches 80%. When the SOH falls below 80%, the simulation ends, and this time atwhich the simulation ends is the battery lifetime (L) in years.

It should be noted that as the cycle-life curves are taken from the manufacturer’sdatasheets, choice of a different battery product or type will impact the lifetime calcula-tions.

5.3.5. POWER MANAGEMENT SCHEME FOR STANDALONE SHS

The algorithm used for power management in every time step (1 minute) is shown inFigure 5.4. Ebatt refers to the battery capacity in Wh in any time step.

As shown in the flowchart, at the start of every time step t , the excess power or the

5

102 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

load deficit is computed as shown in Equation 5.6.

Pexcess = PPV −Pload −Ploss (5.6)

Pdeficit = Pload − (PPV −Ploss)

Where Ploss is the combined power lost in the PV and load converters. If Pexcess > 0,then the load is fully met, and the excess energy either goes to charge to the battery, ordumped if the battery is full.

The case where Pexcess < 0 means that there is an energy deficit and has to be fed fromthe battery. If Ebatt −Ebatt,min ≥ the energy deficit for the time step, then the load is fullymet. In the case where the battery cannot feed the load fully or at all, then the unmetload results in Efail. At the instant where Efail > 0, LLP = 1 for the given time step. It mustbe noted that a C-rate limit of 5C was imposed on the battery operation. This processrepeats for every time step until the one year simulation is over, corresponding to 1 yearof the load data and meteorological data.

5.3.6. CONVERTER RATINGThe power converters shown in Figure 5.2 need to be appropriately rated. Power rating ofeach converter is chosen depending on its intended application as seen below.

PV CONVERTER SIZING

The PV converter optimal size was selected based on the sizing ratio RS defined in Equa-tion 5.7.

RS =PPV,Peak

PNom,conv(5.7)

Where PPV,Peak is the peak PV power and PNom is the rated converter power. Therefore, asizing ratio RS < 1 implies oversizing of the converter.

In the studies conducted in [23, 24], the effect of varying RS on the total yearly en-ergy output from the PV converter was analyzed. The PV array output depends on thegeographical location, and rarely has an output that is equal to or larger than its ratedpeak power. Hence, having RS = 1 or lower for a relatively small time during the year isunnecessary. In this study, the same approach for finding the optimal RS was performed.The yearly energy yield from the PV module was obtained for values of 0.5 ≤ RS ≤ 1.5 asshown in Figure 5.5.

The figure shows that for RS = 0.9, the maximum energy yield is obtained. However,for RS = 1.27, 96% of the maximum yield is obtained. Hence, for an around 41% reductionin the sizing ratio (from 0.9 to 1.27), only 4% of the energy yield is lost. Therefore, an RS of1.27 was selected as the optimal sizing ratio to obtain the PV converter rating PNom,conv,as it is the highest sizing ratio (and therefore most sizing gains) that guarantees more than95% of maximum achievable yield. Moreover, past studies on solar converter sizing haveconcluded that a converter could be undersized by up to 30% of the PV array Wp rating,as undersizing causes an insignificant reduction (≤ 5%) in the total yearly energy yieldcompared to the reduction in its cost [23, 24].

LOAD CONVERTER SIZING

The load converter was sized according to the peak load for each tier. Hence, for the loadconverter, PNom = Pmax, which are the values found in Table 3.4.

5.3. METHODOLOGY

5

103

0.6 0.8 1 1.2 1.4

0.92

0.94

0.96

0.98

1

(1.27,0.96)

(0.9,1)

Sizing ratio RS

No

rmal

ized

year

lyen

ergy

yiel

d[-

]

Figure 5.5: Normalized energy yield VS RS

BATTERY CONVERTER SIZING

The battery converter is a bidirectional converter that processes power by charging ordischarging the battery based on the excess PV generation or greater load demand re-spectively. Hence, it should be sized appropriately to allow the maximum net charge ordischarge power to go in/out of the battery. The PNom for the battery converter is thentaken as the maximum between the Pdeficit and Pexcess values.

5.3.7. MULTI-OBJECTIVE OPTIMIZATION FOR STANDALONE SHS SIZINGOptimal SHS sizing is a complex task that brings to the fore the intricate interplay be-tween the different SHS component sizes and the various system metrics described inSection 5.2.1. For example, a larger battery size can lead in general to longer batterylifetimes (at the cost of increased initial investments), while a smaller battery size willresult in loss of load events (high LLP) as there won’t be enough stored energy to powerthe load in the non-sunlight hours. Similarly, a lower than adequate PV size will result inhigh LLP. However, an indiscriminate increase in PV size will result in high Edump values,which is a waste of energy and should be avoided as much as possible. While an increasein PV or battery size increases the initial system costs, a high Edump value increases thelevelized cost of kWh of the system.

DECISION VARIABLES

For the optimal SHS sizing, 3 variables determine the performance of the system basedon the various metrics. These are: PV size (Wp), battery size (Wh), and converter ratings(W). However, as seen in Section 5.3.6, the converter rating is dependent on the PV sizeand the load. The load profile is a given for the specific SHS application (MTF tier-basedusage). Therefore, the primary independent decision variables used in this study are:

(1) the PV size (Wp) and

5

104 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

(2) the battery size (Wh).

OBJECTIVES

The main objectives can be written as (a) to minimize total cost (b) to minimize LLP and(c) to minimize Edump (or Rdump). However, the total system cost can be considered as asum total of PV cost, battery cost, and the converter cost. Each of these costs are directlyproportional to the rating of the SHS components. Additionally, the battery costs consistof both the initial costs and the replacement costs, owing to the lowest lifetime amongstthe other SHS components. As the replacement costs go down with increasing batterylifetime, minimizing the total battery costs is the same as minimizing the battery sizewhile maximizing the battery lifetime (although battery size and lifetime are not mutuallyindependent variables).

The converter sizes are again dependent on the PV and load and therefore cannotbe independently minimized to lower the costs. PV size is in a way already reflected inminimizing the dumped energy (Edump). Therefore, the objective of minimizing total SHScosts can be crystallized down to the objectives of minimizing battery size, maximizingbattery lifetime while minimizing Edump and LLP . This adaptation of the cost objectiveinto battery size and lifetime helps in avoiding the actual costs in $ ore, for example, andmaintains the generality of the methodology and the results.

Consequently, the following objectives are used for multi-objective optimization inthis study.

(1) to minimize the battery size

(2) to maximize the battery lifetime

(3) to minimize the LLP and

(4) to minimize the Rdump.

CONSTRAINTS

Constraints are necessary in the MOO process to curtail optimization run times andeliminating unwanted navigation of the algorithm employed within the search space. Inthis study, constraints were placed on the LLP and the Edump. Although LLP and Edump

form part of the objective set, the additional constraints augment the efficiency of theoptimization computation. The constraints used in the MOO can be stated as:

(1) LLP ≤ 10%

(2) Edump ≤ yearly load or Rdump ≤ 1.

GENETIC ALGORITHM-BASED MOOFor performing a multi-objective optimization (MOO), the genetic algorithm toolboxof MATLAB was utilized. The gamultiobj function within the GA toolbox of MATLAB isbased on a Non-dominated Sorting GA (NSGA)-II variant [25]. Figure 5.6 outlines thesteps contained in the GA-based MOO used in this study. The functioning of the geneticalgorithm can be explained as follows.

5.3. METHODOLOGY

5

105

Start InitializationSelection based

on fitnessMating

Crossoverand mutation

Fitness of childrenand ranking

ConvergenceEndyes

no

Figure 5.6: Flow chart showing the steps followed in the GA-based MOO.

1. Initialization. The initialization step involves generating a random population of‘N’ individuals which represent the first generation. Each individual has a set ofcharacteristics, which are the PV and battery capacity in Wp and Wh, respectively.For example, for an individual X:

X = [PVX ,B at tX ] (5.8)

The final step of initialization also includes evaluating the fitness.

2. Fitness-based selection. In this step, the fitness of each individual is assessedaccording to the objective functions:

Obj.functions =

Min : LLPMax : LMin : Edump

Min : Ebatt

and the fittest individuals are selected for the mating, while the rest are disregarded.

3. Mating. A pair from the pool of fittest individuals is selected, producing two off-spring individuals that share characteristics from each of the parents. The processis repeated until a new population of ‘N’ individuals is obtained.

4. Crossover and mutation. These two stochastic operators serve to randomly alterthe characteristics of some of the “child” population to maintain some variability inthe algorithm.

5. Fitness of children. The same fitness assessment and selection takes place for thechildren individuals, which represent the new generation.

6. Checking convergence criteria The convergence criteria used are:

(a) Gi =Gmax, where Gi is the i th generation

(b) ∆Si ≤∆S,max for each objective, where∆Si is the spread between the objectivesfor Gi and Gi+1.

If the convergence criteria are met, the optimization process stops. Otherwise, theoptimization process repeats from Step 2 until convergence is reached. In Table 5.1, theparameters and convergence criteria used in this study are shown.

5

106 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

Table 5.1: GA parameters and stopping conditions used in this study.

Parameter Symbol ValuePopulation size Nmax 25Generation limit Gmax 500Spread tolerance ∆S,max 10−4

It must be noted that the a multi-objective optimization need not lead to uniqueoptimization solutions. Instead, a Pareto set of solutions is usually obtained, which(in terms of GA) are the individuals with fitness functions that are non-dominated byany other individual in the search space. Here, dominance refers to the attribute of anindividual to score lower (better) than all other individuals for that particular fitnessfunction. For example, an individual x dominates y in the population when [26]:

fi (x) ≤ fi (y) ∀ i and (5.9)

fi (x) < fi (y) for at least 1 i

5.4. RESULTS AND DISCUSSIONS

5.4.1. DEPENDENCE OF SHS PARAMETERS ON SIZE

As described in Section 5.3.7, there is a complex interplay of the various SHS parameterswith respect to the SHS size. In this section, the impact of sizing on each of the systemmetrics (LLP , Rdump, and lifetime) is presented.

LLPFigure 5.7 presents a contour plot depicting the variations in LLP depending on thePV and battery size for a tier 3 load profile with VRLA battery. It can be seen that thesame LLP can be achieved by either increasing the battery size or the PV size. Moreover,minimal sizing can be achieved if the knee point of the curve is chosen for a given LLP .However, the knee point, or the least distance of the curve from the origin, is a least-costoperating point only if the PV and battery are equally cheap or expensive.

In other words, for the same technology costs ($/Wp and $/Wh), the least-cost operat-ing point is the knee point. Deviations from the knee point to achieve least-cost sizing ispossible if the ratio between the technology costs is precisely known. Additionally, alsoshown in Figure 5.7 are the battery sizes based on the days of autonomy and nights ofautonomy, represented as vertical lines. It can be seen how a “rule of thumb” of 2 DOAusing a “back of the envelope” method to arrive at battery size (Equation 5.4) can result inoversizing of the battery, as even for 1 DOA, Equation 5.4 yields a battery size of 1363 Whfor a VRLA battery assuming an 80% DOD. Using NOA method need not be very preciseeither, and also depends largely on the proportion of the load profile that falls in thenon-sunlight hours. The exact LLP achieved using such methods will also depend onthe operational PV module size. In general, these “rule of thumb”-based approaches canlead to inaccurate battery sizing leading to either an expensive or an inadequately reliablesystem.

5.4. RESULTS AND DISCUSSIONS

5

107

500 750 1000 1250 1500

200

300

400

500

600

1 ·10 −2

2 ·10 −2

5 ·10−2

0.1

0.1

0.2

0.21

DO

A

1N

OA

2N

OA

Battery size [Wh]

PV

size

[Wp

]

Figure 5.7: LLP contours based on PV and (VRLA) battery sizes for a tier 3 load profile.

500 750 1000 1250

200

300

400

500

5 ·10−2 5 ·10−20.10.1

0.20.2

0.30.3

0.5 0.5

0.75 0.75

1 1

1.25 1.25

Battery size [Wh]

PV

size

[Wp

]

Figure 5.8: Rdump contours based on PV and (VRLA) battery sizes for a tier 3 load profile.

DUMP RATIO

Figure 5.8 presents a contour plot depicting the variations in Rdump depending on thePV and battery size for a tier 3 load profile with VRLA battery. As seen in the figure, theRdump increases with increasing PV size. To an extent the increase in battery size helpsin reducing the dump ratio. However, after around 900 Wh, the increase in battery isalmost irrelevant for the considered PV and battery size range. It can already be seen thatan interesting trade-off emerges between LLP and Rdump. For example, to maintain an

5

108 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

LLP of 0.05 and below (Figure 5.7), almost 300 Wp of PV module is necessary. On theother hand, a 300 Wp of PV module size will guarantee an Rdump of at least 0.3 as seen inFigure 5.8.

BATTERY LIFETIME

The trade-off between LLP and Rdump becomes more intricate when the battery lifetimeis considered into the mix. Figure 5.9 presents a contour plot depicting the variations inbattery lifetime (L) in years depending on the PV and battery size for a tier 3 load profilewith VRLA battery.

500 750 1000 1250 1500

200

300

400

500

6 7 8 9

10

12

Battery size [Wh]

PV

size

[Wp

]

Figure 5.9: Lifetime contours in years based on PV and (VRLA) battery sizes for a tier 3 load profile.

For the same PV size, the battery lifetime increases with increasing battery size. For thesame battery size, the lowest PV size leads to relatively lowest battery activity and thereforehighest lifetime. As the PV size increases for the same battery size, the battery lifetimeinitially decreases due to increasing battery activity contributing to higher cyclic ageing.However, at the higher end of the PV sizing range, the battery is being operated largelyat relatively lower DOD levels, leading to relatively higher lifetimes than the medium PVrange.

5.4.2. MULTI-OBJECTIVE OPTIMIZATION FOR SHS SIZING

Based on the methodology described in Section 5.3.7, a multi-objective optimization wasperformed for the various objectives. Consequently, different Pareto fronts are obtainedthat give the optimal PV and battery sizes that dominate at least one objective functionwithout being worse off in the other objectives, as shown in Equation 5.9. Figure 5.10shows for the tier 3 case the Pareto set of points for LLP , Rdump, which also performoptimally with respect to the other objectives of lifetime and battery size. As expected,the solution space is bounded by the constraints specified in Section 5.3.7.

5.4. RESULTS AND DISCUSSIONS

5

109

0.2 0.4 0.6 0.8 10.00

0.02

0.04

0.06

0.08

0.10

B

A

Rdump [-]

LL

P[-

]

Figure 5.10: LLP VS Rdump for a Pareto set of points for the tier 3 case. Points A and B perform differently on the4 objectives, but still form a part of the Pareto set of system solutions.

6 8 10 120.00

0.02

0.04

0.06

0.08

0.10 B

A

Lifetime [years]

LL

P[-

]

Figure 5.11: LLP VS lifetime for a Pareto set of points for the tier 3 case. Points A and B perform differently onthe 4 objectives, but still form a part of the Pareto set of system solutions.

Figure 5.11 and Figure 5.12 depict the same Pareto set of points for the system metricsof LLP and lifetime, and lifetime and Rdump, respectively. Some of the points that weretoo close to each other have been removed for clarity. It can be seen that the varioussystem sizes represented by the points in the Pareto set perform differently across thevarious system metrics. As long as the different objectives are not weighted, the appliedmethodology does not favour a particular system size over another. Therefore, the task of

5

110 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

0.2 0.4 0.6 0.8 1

6

8

10

12

B

A

Rdump [-]

Life

tim

e[y

ears

]

Figure 5.12: Lifetime VS Rdump for a Pareto set of points for the tier 3 case. Points A and B perform differently onthe 4 objectives, but still form a part of the Pareto set of system solutions.

optimal SHS sizing for a particular tier will depend on selecting an optimal size based onadditional filters and categorization, as described below in Section 5.4.2.

It should be noted that the GA-based MOO chooses the Pareto front based on theperformance across all 4 objective functions and not just the 2 functions described inthese figures. Therefore, each figure shows the Pareto points that satisfy not only the twoobjective functions represented on the x and y-axes, but all four objective functions statedin Section 5.3.7. This is illustrated with the help of 2 labeled Pareto points A and B inFigures 5.10 to 5.12. Point A satisfies a very high battery lifetime as well as a low Rdump

(11.93 years and 0.1 respectively), but shows a relatively high LLP (0.07). Furthermore,the battery size, which is also an objective to be minimized in the optimization, is notexplicitly represented in the Pareto fronts in Figures 5.10 to 5.12, but is playing an equallyimportant role in the selection of optimal Pareto points. For example, point A leads toa storage size of 1800 Wh for tier 3 SHS. On the other hand, point B, which seems to bespecifically non-dominant for the LLP , Lifetime and Rdump simultaneously (0.095, 5.81,and 0.81 respectively), actually has the lowest battery size (720 Wh), and is therefore partof the Pareto front.

OPTIMAL SHS SIZES FOR TIERS OF THE MTFSimilar to the results shown in Figure 5.10 to Figure 5.12, the MOO was performed forall the tiers of the MTF. To determine the optimal system sizes per tier, a selection has tobe made from the Pareto set of points allowing relative trade-offs. Firstly, three differentcategories of LLP are defined, viz., (a) LLP ≤ 0.1, (b) LLP ≤ 0.05, and (c) LLP ≤ 0.02.Secondly, the smallest battery size that satisfies the above LLP criteria in each categoryare chosen. This gives rise to unique PV-battery size combinations per tier. Accordingly,the final optimal SHS sizes for every tier of the MTF is shown in Table 5.2. The table also

5.4. RESULTS AND DISCUSSIONS

5

111

Table 5.2: Optimal SHS sizes that satisfy the different LLP criteria, along with the required converter sizes andthe performance of these system sizes on the system metrics.

LLP [-]Tier PV

(Wp)Battery[Wh]

Converter [W] Lifetime Rdump LLP

PV Load Battery [Years] [-] [-]

≤ 0.1

1 20 60 15 12 15 6.2 0.82 0.0472 70 210 53 51 52 5.9 0.48 0.13 380 720 285 154 259 5.8 0.81 0.14 1620 2520 1215 1670 1660 5.4 0.89 0.0995 4050 5300 3038 3081 2961 5.7 0.99 0.089

≤ 0.05

1 20 60 15 12 15 6.2 0.82 0.0472 80 240 60 51 59 6.3 0.65 0.0393 340 860 255 154 229 6.0 0.56 0.0424 1500 2880 1125 1670 1661 5.7 0.73 0.0465 3800 6150 2850 3081 2777 6.1 0.84 0.0431

≤ 0.02

1 20 70 15 12 15 6.8 0.80 0.0192 90 290 68 51 67 7.5 0.85 0.0193 420 1020 315 154 288 6.7 0.92 0.0124 1740 3560 1305 1670 1660 7 1.0 0.0175 4000 6600 3000 3081 2924 6.45 0.93 0.029

shows the system metrics for the particular PV and battery size along with the convertersizes.

Based on the methodology followed in Section 5.3.6, the converter sizes follow the PVrating and load profile. The PV and battery sizes, however, are a direct result of selectionof a particular PV-battery combination from the Pareto set. Based on the LLP threshold,the selection of the lowest battery size to meet the LLP criteria gives an interesting mixof system sizes. For instance, as seen in Table 5.2, the LLP ≤ 5% criterion sees lower PVand higher battery sizes as compared to LLP ≤ 10% for tiers 3 to 5. This leads to relativelylower Rdump values.

THE LIMITS OF STANDALONE SHSAs seen in Table 5.2, the PV and battery sizes increase drastically between the tiers,especially when going towards tiers 4 and 5. Additionally, it can be seen that no optimalsolutions exist with the given constraints for tier 5 for LLP ≤ 2%. This is because larger-sized SHS that could potentially satisfy the LLP criterion would still end up compromisingthe dump criterion while also leading to extremely high battery sizes. This clearly showsthat there is a limit to the level of electricity access a standalone SHS can provide.

For tiers 1 to 3, a small increment in system size (due to climbing up the tiers forinstance) can be easily achieved, especially in a way such that low LLP values are guar-anteed. However, tiers 4 and 5 require significant increase in system sizing, and withoutthe kind of reliability (low LLP) achievable at the lower tiers. Additionally, the presenceof high power loads also increases the required converter ratings in tiers 4 and 5. If the

5

112 5. EXPLORING SHS BOUNDARIES FOR ELECTRIFICATION: OPTIMAL SHS SIZING

LLP threshold is further lowered to ≤ 1%, even tier 4 has no optimal solution based on thePareto points. The lowest achievable LLP for tier 4 within the given constraints is 1.5%.

STATE-OF-THE-ART SHSIt should be noted that the concept of LLP as a vital parameter in system sizing is usuallynot followed in state-of-the-art SHS. For example, present-day SHS capable of poweringup to tier 2 level loads, are sized from 50 Wp, 300 Wh (for an SHS operational in EastAfrica) to 100 Wp, 1 kWh (for an SHS operational in Cambodia). While this may leadto suboptimal use of the SHS components of PV and battery, having a sizing approachwithout a reasonable approximation of LLP targets for standalone SHS catering to highertiers could amplify the suboptimal usage.

Additionally, as most deployed state-of-the-art SHS are rated less than 200 Wp, thereis a long way to go if SHS are to enable higher tiers of electrification as ubiquitously asthey have been effective with tiers 1 (as pico-solar products) and 2. While tier 3 levelelectrification still seems to be within reach of state-of-the-art SHS, tier 4 and tier 5 levelelectrification demand a much bigger expansion from the standalone SHS, which may notbe practical to implement when additional aspects like financial viability are consideredin off-grid contexts.

ANOTHER APPROACH TO CLIMBING UP THE ELECTRIFICATION LADDER

In the absence of a central grid connection in many of these off-grid regions where SHSare currently deployed, the best way to achieve tier 4 or tier 5 level of electricity access isvia a microgrid. However, a centralized islanded microgrid with central PV and storagerequires high CAPEX investments, which is also why standalone SHS have seen far greaterproliferation than centralized rural microgrids. Moreover, the level of electrification isoften a dynamic requirement, as the electricity needs keep increasing with time [27, 28].In such a case, it is more practical to envisage a bottom-up SHS-based microgrid borneout of the interconnection of standalone SHS, as discussed earlier in Chapter 2. Thiswould enable the sharing of excess energy between the households while improving theoverall system metrics and potentially reducing the battery storage size.

5.5. CONCLUSIONThis study detailed an extensive methodology for the optimal sizing of SHS for every tierof the multi-tier framework for household electricity access. The genetic algorithm-basedmulti-objective optimization performed gave insights on the delicate interdependenciesof the various system metrics on the SHS sizing. Moreover, meeting the energy demandof higher tiers, especially tier 5 is shown to be untenable with purely standalone SHS.Optimal system sizes for each tier of the MTF are presented, and the implications ofthese system sizes are discussed from the perspective of state-of-the-art SHS. Finally,an SHS interconnection-based microgrid is proposed as a potential means to climb upthe so-called electrification ladder, especially to easily enable tier 4 and tier 5 levels ofelectricity access. The work presented in this chapter is expected to shed light on thecomplex, multi-dimensional issue of electrification from the point of view of technicalsystem design while exploring the intricate interdependencies of SHS parameters.

REFERENCES

5

113

The optimal SHS sizes obtained for the various LLP thresholds serve as a reference forquantifying the benefits of a decentralized, interconnected SHS-based microgrid on thesystem metrics in Chapter 6.

REFERENCES[1] N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and

M. Zeman, “Estimating battery lifetimes in solar home system design using a practi-cal modelling methodology,” Applied Energy, vol. 228, pp. 1629 – 1639, 2018.

[2] N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P. Bauer, and M. Zeman,“A simple methodology for estimating battery lifetimes in solar home system design,”in 2017 IEEE AFRICON, pp. 1195–1201, Sept 2017.

[3] V. Vega-Garita, D. D. Lucia, N. Narayan, L. Ramirez-Elizondo, and P. Bauer, “Pv-battery integrated module as a solution for off-grid applications in the developingworld,” in 2018 IEEE International Energy Conference (ENERGYCON), pp. 1–6, June2018.

[4] N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Zeman, “Amodeling methodology to evaluate the impact of temperature on solar home systemsfor rural electrification,” in 2018 IEEE International Energy Conference (ENERGY-CON), pp. 1–6, June 2018.

[5] N. Narayan, Z. Qin, J. Popovic, P. Bauer, and M. Zeman, “Evaluating the techno-economic feasibility of battery technologies in the context of solar home systems,”in 2018 20th European Conference on Power Electronics and Applications (EPE’18ECCE Europe), pp. P.1–P.10, Sept 2018.

[6] O. Ekren and B. Y. Ekren, “Size optimization of a pv/wind hybrid energy conversionsystem with battery storage using response surface methodology,” Applied Energy,vol. 85, no. 11, pp. 1086 – 1101, 2008.

[7] R. Hosseinalizadeh, H. S. G, M. S. Amalnick, and P. Taghipour, “Economic sizing ofa hybrid (pv–wt–fc) renewable energy system (hres) for stand-alone usages by anoptimization-simulation model: Case study of iran,” Renewable and SustainableEnergy Reviews, vol. 54, pp. 139 – 150, 2016.

[8] F. Benavente-Araoz, A. Lundblad, P. E. Campana, Y. Zhang, S. Cabrera, and G. Lind-bergh, “Loss-of-load probability analysis for optimization of small off-grid pv-batterysystems in bolivia,” Energy Procedia, vol. 142, pp. 3715 – 3720, 2017. Proceedings ofthe 9th International Conference on Applied Energy.

[9] S. Chalise, J. Sternhagen, T. M. Hansen, and R. Tonkoski, “Energy management ofremote microgrids considering battery lifetime,” The Electricity Journal, vol. 29, no. 6,pp. 1 – 10, 2016.

[10] M. Bortolini, M. Gamberi, A. Graziani, and F. Pilati, “Economic and environmental bi-objective design of an off-grid photovoltaic–battery–diesel generator hybrid energysystem,” Energy Conversion and Management, vol. 106, pp. 1024 – 1038, 2015.

5

114 REFERENCES

[11] K. Jäger, O. Isabella, A. H. Smets, R. Van Swaaij, and M. Zeman, Solar Energy: Thephysics and engineering of photovoltaic conversion, technologies and systems. UITCambridge, 2016.

[12] I. Kougias, S. Szabó, F. Monforti-Ferrario, T. Huld, and K. Bódis, “A methodology foroptimization of the complementarity between small-hydropower plants and solarpv systems,” Renewable Energy, vol. 87, pp. 1023 – 1030, 2016. Optimization Methodsin Renewable Energy Systems Design.

[13] W. D. Kellogg, M. H. Nehrir, G. Venkataramanan, and V. Gerez, “Generation unitsizing and cost analysis for stand-alone wind, photovoltaic, and hybrid wind/pvsystems,” IEEE Transactions on Energy Conversion, vol. 13, pp. 70–75, March 1998.

[14] R. Ayop, N. M. Isa, and C. W. Tan, “Components sizing of photovoltaic stand-alonesystem based on loss of power supply probability,” Renewable and Sustainable EnergyReviews, vol. 81, pp. 2731 – 2743, 2018.

[15] T. Khatib, A. Mohamed, and K. Sopian, “A review of photovoltaic systems size opti-mization techniques,” Renewable and Sustainable Energy Reviews, vol. 22, pp. 454 –465, 2013.

[16] M. D. Al-Falahi, S. Jayasinghe, and H. Enshaei, “A review on recent size optimizationmethodologies for standalone solar and wind hybrid renewable energy system,”Energy Conversion and Management, vol. 143, pp. 252–274, 2017.

[17] E. De Schepper, S. Lizin, B. Durlinger, H. Azadi, and S. Van Passel, “Economic andenvironmental performances of small-scale rural pv solar projects under the cleandevelopment mechanism: the case of cambodia,” Energies, vol. 8, no. 9, pp. 9892–9914, 2015.

[18] M. Paulitschke, T. Bocklisch, and M. Böttiger, “Sizing algorithm for a pv-battery-h2-hybrid system employing particle swarm optimization,” Energy Procedia, vol. 73,pp. 154–162, 2015.

[19] F. A. Bhuiyan, A. Yazdani, and S. L. Primak, “Optimal sizing approach for islandedmicrogrids,” IET Renewable Power Generation, vol. 9, no. 2, pp. 166–175, 2015.

[20] Meteotest, “(software) meteonorm ver 7.1,” 2014.

[21] N. Narayan, “Electrical power consumption load profiles for households with dcappliances related to multi-tier framework of household electricity access.” https://doi.org/10.4121/uuid:c8efa325-87fe-4125-961e-9f2684cd2086, 2018.

[22] “Installation , commissioning and operating instructions for valve-regulated station-ary lead-acid batteries - solar battery data sheet,” 2015.

[23] X. Camps, G. Velasco, J. de la Hoz, and H. Martín, “Contribution to the pv-to-invertersizing ratio determination using a custom flexible experimental setup,” AppliedEnergy, vol. 149, pp. 35 – 45, 2015.

REFERENCES

5

115

[24] S. Islam, A. Woyte, R. Belmans, and J. Nijs, “Undersizing inverters for grid connection-what is the optimum?,” Proceedings of PV in Europe, pp. 780–783, 2002.

[25] MATLAB, gamultiobj Algorithm. Natick, Massachusetts: The MathWorks Inc., 2018.

[26] D. Kalyanmoy, Multi objective optimization using evolutionary algorithms. JohnWiley and Sons, 2001.

[27] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrifica-tion,” in 2017 International Conference on the Domestic Use of Energy (DUE), pp. 8–15,April 2017.

[28] N. Narayan, Z. Qin, J. Popovic-Gerber, J.-C. Diehl, P. Bauer, and M. Zeman, “Stochasticload profile construction for the multi-tier framework for household electricityaccess using off-grid dc appliances,” Energy Efficiency, Nov 2018.

6QUANTIFYING THE BENEFITS OF

SHS-BASED DC MICROGRIDS

There is hopeful symbolism in the fact that flags do not wave in a vacuum

Arthur C. Clarke

ABSTRACTAs seen in Chapter 5, state-of-the-art SHS can only provide electricity access with ad-equate power supply availability up to tier 2, and to some extent, tier 3 levels of theMulti-tier Framework (MTF) for measuring household electricity access. When consider-ing system metrics of loss of load probability (LLP) and battery size, meeting the electricityneeds of tiers 4 and 5 is untenable through SHS alone. Alternatively, a bottom-up micro-grid composed of interconnected SHS is proposed in this chapter. Such an approach canenable the so-called climb up the rural electrification ladder. The impact of the microgridsize on the system metrics like LLP and energy deficit is evaluated. Finally, it is found thatthe interconnected SHS-based microgrid can provide more than 40% and 30% gains inbattery sizing for the same LLP level as compared to the standalone SHS sizes for tiers5 and 4 of the MTF, respectively, thus quantifying the definite gains of an SHS-basedmicrogrid over standalone SHS. This study paves the way for visualizing SHS-based ruralDC microgrids that can not only enable electricity access to the higher tiers of the MTFwith lower battery storage needs but also make use of existing SHS infrastructure, thusenabling a technologically easy climb up the rural electrification ladder.

OUTLINEThis chapter is divided into five sections. Section 6.1 introduces the chapter and presentsa literature review. Section 6.2 describes in detail the methodology used in the study.

This chapter is based on: N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, & M.Zeman. (2019). Quantifying the Benefits of a Solar Home System-Based DC Microgrid for Rural Electrification.Energies, 12(5), 938.

117

6

118 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

Section 6.3 discusses the results and their implication while quantifying the benefitson the various system metrics of going from standalone SHS to an SHS-interconnectedmicrogrid. Finally, Section 6.4 concludes the chapter with the closing thoughts on theinterconnection benefits and recommendations for future work.

6.1. INTRODUCTIONIn this chapter, an alternative, bottom-up, SHS-based microgrid is proposed as a solutionto achieve higher tiers of electrification, and the benefits of such an interconnectedSHS-based microgrid are quantified over standalone SHS.

6.1.1. LITERATURE STUDY

INADEQUACY OF CONTEMPORARY SHS

Most plug-and-play SHS for households currently being offered in the off-grid marketsegment belong to the 11 Wp to 100 Wp category [1, 2]. For locations with 4 or moreequivalent sun hours per day of solar insolation (a common attribute for locations in theequatorial and tropical areas), such a PV rating can easily satisfy the electricity needs upto tier 2 (Table 2.1), provided adequate battery storage is present in the system. However,there is a scarcity of SHS-based solutions that cater to higher tiers of the MTF. For instance,tier 4 requirements are stated to be more than 3.4 kWh per day, which requires more than800 Wp of rated PV in the SHS, which is a far cry from contemporary SHS ratings availablein the field (typically around 100 Wp or less)[1, 3]. The study presented in Chapter 5on optimal sizing of SHS while increasing the battery lifetime, decreasing the excessunused energy, and reducing the LLP concluded that standalone SHS are not tenable tobe used for tier 4 and tier 5 level electrification. In fact, the lowest LLP attainable by anoptimally sized SHS catering to tier 5 electricity demand was seen to be 0.029 [3], whileseveral off-grid projects keep 0.02 as a lower limit of the LLP [4, 5]. For off-grid systems,typical LLP values based on the specific application can be 0.01 and 0.0001 for domesticillumination and telecommunications respectively [6]. Moreover, due to the mismatchbetween the electricity demand and production, which is to some extent abated with thehelp of battery storage, there is usually some amount of energy that is wasted, leading tounderutilization of the PV-generated energy. This utilized energy in standalone SHS inliterature has been seen to be as much as 31%, for simulated SHS installed in Bangladesh[7]. For SHS installed in Rwanda, around 65% of the generated energy remains unutilized[8]. The SHS considered in these studies are currently catering to tier 1 and tier 2 usersonly. For higher tier usage, the scaling up of PV and battery size needed to meet the energydemand would mean that the unutilized component of generated energy also scales upin magnitude, thus leading to more "wasted energy". Even though contemporary SHSmostly cater to tier 1 and tier 2 usage, the climb up the electrification ladder is inevitable.Hence, there is a need to also focus on the larger goal of being able to satisfy the highertier energy demands. As contemporary standalone SHS suffer from the aforementioneddrawbacks, a microgrid-based electrification needs to be explored, especially for meetingthe electricity demands of the higher tiers. This is because microgrids can potentiallycater to much higher power needs as opposed to single SHS with only one PV panel atthe household level. Furthermore, grid-extension, which is the third option apart from

6.1. INTRODUCTION

6

119

standalone systems and microgrids, has not been considered in this chapter as for themajority of the unelectrified population it is not a cost-effective solution, as also discussedin Chapter 2.

RURAL MICROGRIDS

In the absence of a central grid connection in many of the un(der)-electrified off-gridregions where SHS are currently deployed, the best way to achieve tier 4 or tier 5 levelof electricity access is via a microgrid. In the past, rural microgrids in literature haveencompassed different categorizations: centralized [9] and decentralized microgrids[10]; grid-connected [11] and islanded (or off-grid) [12]; AC [13], DC [14] and hybrid [15];Renewable energy-based [16] and hybrid energy-based (with diesel) [17]. However, giventhe falling PV prices, solar-based solutions like pico-solar products and SHS have alreadyseen a very high rate of proliferation and acceptance. Additionally, the SHS available inthe field are mostly DC-based, i.e. without AC components. Therefore, the focus of thisstudy is mainly on islanded solar-based rural DC microgrids. Such a microgrid can beimplemented in three ways, viz.,

1. a centralized microgrid, with central PV generation and centralized storage;

2. a semi-decentralized microgrid, with central PV generation and decentralizedstorage, or vice-versa;

3. a fully decentralized microgrid, with decentralized PV generation, decentralizedstorage, and DC distribution.

Compared to standalone SHS at the household level, a centralized islanded microgridwith central PV and storage requires high CAPEX investments, and few private playerswant to set up such a centralized endeavour with the high risks currently associatedwith such an investment. Semi-decentralized microgrids are also being explored, witha central village solar kiosk, with individual households having some loads and a basicenergy storage unit like battery. This can sometimes also be implemented in the formof a single entrepreneur operating the kiosk and leasing out batteries that are chargedfrom the centralized PV [18]. This does not strictly qualify as a microgrid, as there isno distribution grid present. Additionally, semi-decentralized solutions have also beenlimited by the extent of the electricity they are able to provide.

BOTTOM-UP, DECENTRALIZED INTERCONNECTED SHS-BASED RURAL MICROGRIDS

In general, the level of electrification is often a dynamic requirement, as the electricityneeds keep increasing with time [19, 20]. For example, households might not need adirect tier 4 or tier 5 connection but might move from tier 3 to tier 4 over a span oftime. This would mean that the planned microgrid has to expand and scale up withthe increase in the energy demand. In such a case, it is more practical to envisage abottom-up SHS-based microgrid born out of the interconnection of standalone SHS.This would lead to a reduced storage requirement as the households will be able todepend on the excess generation of other households in the microgrid network. Moreover,since the lower tier electrification is already underway through SHS in large parts of theworld, an interconnected SHS-based decentralized rural DC microgrid is an option that

6

120 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

is easier to retrofit when visualizing future rural microgrids. The advantages of such aninterconnected SHS-based microgrid has already been discussed in Section 2.6.1.

The idea of bottom-up, decentralized rural DC microgrids that can grow organicallywith time has seen more proponents in literature [21, 22]. However, existing studies like[23, 22] mainly identify the financial benefits of interconnecting SHS. The advantage ofdemand diversity, that is the load profile peaks of different households occurring at differ-ent points in time, due to SHS interconnection, is captured in [8]. Moreover, these studies,based on existing SHS, are limited to tiers 1 to 2 with 12 V battery and distribution. Scalingup the energy ladder would demand high power loads, and therefore a much highervoltage for DC distribution. Additionally, higher power loads that enable productive useof energy are not considered in these studies. Moreover, there is insufficient informationin literature on the quantified benefits of microgrid over SHS in terms of system sizingand other system metrics like power supply availability and excess energy.

Therefore, this chapter focuses on the concept of bottom-up SHS-based microgridsthat can grow organically based on the electricity demand and the affordability of thehouseholds in a given location. Furthermore, this chapter endeavours to quantify thebenefits of such an interconnected SHS-based microgrid over standalone SHS.

6.1.2. CONTRIBUTIONS OF THIS CHAPTER

Following are the main contributions of this chapter.

• A bottom-up, organically growing microgrid is modeled that enables climbing upthe rural electrification ladder through energy sharing.

• The benefits of SHS interconnectivity over standalone SHS for enabling higher tiersof electricity access are quantified in the form of system metrics of storage size, lossof load probability and excess energy.

• A modular SHS-based architecture is proposed that can not only enable modularintra-household expansion of the SHS but also allow for inter-household scalabilityof a meshed DC microgrid.

6.2. METHODOLOGY

6.2.1. LOCATION AND METEOROLOGICAL INPUTS

Meteonorm software database was used to obtain the meteorological data of a city inIndia (that enjoys 5 equivalent sun hours of irradiance per day) as the ground levelmeteorological data was not readily available for the remote rural areas [24]. This datacan be assumed to be representative of other areas in similar latitudes, as most of theun(der)-electrified areas lie in the tropical latitudes enjoying comparable equivalent sunhours. The meteorological data is composed of ambient temperature, wind speed, andground level irradiance, which are all required for modelling the dynamic PV output forSHS. This has been extensively discussed in Chapter 5. The data used in this study alsohas a resolution of 1 minute.

6.2. METHODOLOGY

6

121

6.2.2. STOCHASTIC LOAD PROFILES

Similar to Chapter 5, stochastic load profiles were used as inputs in this study for thevarious levels of MTF. The construction of these household level load profiles has beendetailed in Chapter 3 and published in [20]. Figure 6.1 shows sample load profiles from arepresentative day for tier 4 and tier 5 level electrified households. The household-levelload profiles for all the tiers differ on a day-to-day basis; the datasets have been obtainedfrom [25].

0 4 8 12 16 20 240

500

1,000

1,500

2,000

Time [h]

Pow

er[W

]

Tier 4Tier 5

Figure 6.1: 1-day load profiles for tier 4 and tier 5 levels of electricity consumption. Data sourced from [25].

As the focus of the work is to determine the adequacy, and quantify the benefits of, anSHS-based microgrid for higher tiers of electrification as compared to standalone SHS,only tiers 4 and 5 have been considered in this study while modelling the energy exchangebetween the SHS in the microgrid.

6.2.3. SYSTEM METRICS AND PARAMETERS

The system metrics used in this work are discussed below.

LOSS OF LOAD PROBABILITY (LLP)This has been already discussed in Section 5.2.1. In Chapter 5, different LLP thresholdswere investigated for SHS applications.

UNSATISFIED LOAD ENERGY (Efail)This has also been already discussed in Section 5.2.1.

ENERGY DUMP (Edump)This has also been already discussed in Section 5.2.1.

6

122 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

PER HOUSEHOLD METRICS

To translate the SHS level metrics to the microgrid level, certain per household metricsare defined as the average metric per household in the microgrid. Specifically, these are:

• Average LLP per household:

LLP =∑H

i LLP

H(6.1)

where H is the total number of households in the microgrid.

• Average Rdump per household:

Edump =∑H

i Edump

H(6.2)

• Average Efail per household:

Efail =∑H

i Efail

H(6.3)

6.2.4. OPTIMAL STANDALONE SHS SIZES FOR THE MTFBased on the multi-objective optimization for SHS sizing described in Chapter 5, optimalSHS sizes were identified as described in Section 5.4.2. Table 5.2 captures these optimalSHS sizes, which will serve as the reference for comparing the performance of the in-terconnected SHS-based microgrid across the same system metrics on a per householdbasis.

As seen in Table 5.2, tiers 4 and 5 required very high sizes of standalone SHS batterysizes. No standalone SHS size could satisfy the LLP threshold of ≤ 2% for tier 5 within thegiven constraints of the study [3].

6.2.5. SHS INTERCONNECTION-BASED DC MICROGRIDAn interconnected SHS-based DC microgrid is expected to enable the excess energysharing between the households while improving the overall system metrics of LLP andEdump. This section details the steps involved in modeling the energy exchange betweendifferent SHS in such a microgrid. A concept illustration of such an interconnected SHS-based microgrid was shown in Figure 2.12. One of the houses in the figure is shownwithout an installed SHS (illustrated as a non-PV rooftop on the bottommost house),indicating that "load only" consumptive households can also be potentially part of such amicrogrid to leverage the excess energy being produced from the PV of other SHS in themicrogrid.

MODULAR SHS-BASED MICROGRID ARCHITECTURE

Climbing up the so-called rural electrification ladder requires the architecture to supportthe expansion of the off-grid system to cater to increased load demand. On the otherhand, the architecture should also allow for easy interconnectivity with other SHS to forma meshed DC microgrid. Therefore, a modular architecture for SHS is proposed that couldpotentially allow for expansion of the system at the household level, while also enablingscaling up of the microgrid size. Figure 6.2 shows the modular architecture.

6.2. METHODOLOGY

6

123

SHS microgrid

PV

BatteryLoads

small

Loads

350 V

Battery

bus

DC48 V 350 V

Figure 6.2: Modular DC architecture for an SHS-based DC microgrid that enables intra-household growthoption as well as scalability in the form of a meshed DC microgrid. Dashed battery with converter illustratesthe modular capability. Similar modular expandability can be considered for loads and PV too. The additionalhardware needed to go from standalone SHS to microgrid, along with the high power loads at 380V that themicrogrid can support, are denoted in blue colour.

All the SHS components and DC loads are connected to the central DC bus via con-verters. The SHS can be modularly expanded as shown in Figure 6.2 with the dashedbattery at the bottom. The DC bus is assumed to be operated at a reference of 48 V, whichhas been tested in rural DC microgrid pilot in the past [27]. For the higher power loads,whether within a household (for tier 4/tier 5) or more communal loads like solar pumps,the higher voltage line is available that is operated at a reference of 350 V. This helps tokeep the current and therefore cable losses low. Additionally, such an architecture alsohelps in scaling up this meshed DC microgrid, where more of such SHS of different sizescould be interconnected. It is assumed that the power exchange capacity of such anarchitecture is only limited by the power demand of the high power loads on the 350 Vbus.

POWER MANAGEMENT SCHEME FOR INTERCONNECTED SHS-BASED MICROGRID ARCHI-TECTURE

The detailed PV and storage modeling at every SHS level has been covered extensivelyin previous works by the authors in [28, 3, 26]. The power management scheme imple-mented at the SHS level was shown in the flowchart in Figure 5.4.

6

124 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

Note that a fully decentralized microgrid architecture will also benefit from a decentral-ized control scheme at the SHS level . While going from SHS level to the interconnectedSHS microgrid, additional information needs to be considered in every time step. Fig-ure 6.3 shows the algorithm for the interconnected power management scheme. Thesame algorithm that occurs at every time step t in Figure 5.4 takes place in this scenario,and is repeated for each of the total number of households ‘H’. In this step, the LLP ,Edump, Efail and Ebatt for each household are calculated.

Start

Inputs

LLP , Edump, Efail

and Ebatt foreach household

at time step

Minimize Efail andcharge batteries

with energy surplus(Edump & Ebatt)

New LLP , Edump,Efail and Ebatt at t

t ≤ T?

End

t=

t+1

Yes

No

Figure 6.3: Steps involved in the modeling of energy sharing between the SHS for a given number of households.T denotes the total time period of simulation. Apart from minimizing Efail, the shaded block involves sharingenergy surplus, as detailed in Figure 6.4.

The most important step in this algorithm is the energy sharing process that occurstowards the end of the time step, as indicated by the shaded process block in the flowchart

While this is not the focus of this chapter, a decentralized control scheme for DC-interconnected SHS hasbeen discussed in Appendix C

6.2. METHODOLOGY

6

125

Edump, LLP,Ebatt, Efail at t

∑Edump ≥ 0 &

∑Efail ≥ 0 ?

Supply energydeficit to min-

imize Efail

Efail = 0 ?

Ebatt ≥ Ebatt,min

Edump > 0?

New LLP=1 foreach householdwith Efail 6= 0

Charge batterieswith excess Edump

Supply energyto minimize Efail

Updated Edump,LLP, Ebatt,Efail at t

Input

Output

No

No

Yes

YesYes

Yes

No

No

Figure 6.4: Algorithm for sharing the excess energy based on the ranking of load deficit (in terms of Efail andbattery depth of discharge (DOD)).

in Figure 6.3. During this step, the energy sharing between the households takes placeto balance the overall energy surplus and deficit in the microgrid. The entire process isdetailed in Figure 6.4.

First, the LLP , Efail and Edump for each household are taken as inputs. The total dumpobtained is used as the first measure to eliminate the energy deficit of the households inan ascending order as shown in Equation 6.4.

Efail,t = [Efail,1,t ,Efail,2,t , ....Efail,H ,t ] (6.4)

where Efail,t is the ordered array containing the energy deficit per household for time stept , Efail,i ,t is the energy deficit in time step t for the i th household, and Efail,i ≤ Efail,i+1.Hence, the household with the lowest energy deficit is the first to be supplied from theenergy dump, and so on for the remaining households. Prioritizing the households with alow Efail minimizes the number of households with an Efail > 0 for time step t . This resultsin reducing the average LLP for the SHS community in the microgrid.

If there are still more households left with energy deficit, then the SHS batteries of therest of the microgrid come into play. This happens when Edump,t = 0 &

∑Hi=1 Efail,i ,t > 0.

6

126 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

The available battery storage is then sorted in a descending order, where the SHS with thehighest available battery energy is the first to supply energy to minimize the deficit.

Ebatt,t = [Ebatt,1,t ,Ebatt,2,t , ...Ebatt,H ,t ] (6.5)

where Ebatt,t is the matrix containing the battery per household for time step t , Ebatt,i ,t

is the battery energy in time step t for the i th household, and Ebatt,i ,t ≥ Ebatt,i+1,t . Theenergy deficit minimization process ends when

1.∑H

i=1 Efail,i ,t = 0 or

2. Ebatt,i = Ebatt,min,i ∀ Ebatt,i

After this process, any household that still has an energy deficit has an LLP = 1 for thattime step. In the case where all the energy deficit is satisfied but some excess energy stillremains from Edump, the batteries in the microgrid are recharged. This can be done inmultiple ways. In this study, 3 different excess energy sharing mechanisms are exploredfor charging the batteries with excess energy.

A DOD-BASED PROPORTIONAL EXCESS ENERGY SHARING

Echarge,i ,t = xi ,t .H∑

i=1Edump,i ,t (6.6)

Where xi ,t =DODi ,t∑H

i=1 DODi ,t(6.7)

Where xi ,t is the share of i th household out of the remaining surplus energy, Edump,i ,t

is the excess energy available from the i th household at time step t , and Echarge,i ,t is the

energy allocated for sharing the battery of the i th household at time step t . Hence, thehousehold with the highest DOD obtains the highest share from the available combinedexcess energy.

The concept of depth of discharge (DOD)-based proportional excess energy sharingcan be best illustrated in the following example. Suppose there are 4 SHS (SHS1, SHS2,SHS3, and SHS4) with respective battery DOD levels of 10%, 20%, 30%, and 40% at timet wherein the individual load deficits have already been already taken care of. Supposethere is still a total excess Edump of 1 kWh available, while each battery has a rated capacityof 2 kWh. Consequently, through the DOD-based, proportional excess energy sharingmethod described in Equations 6.6 and 6.7, the 4 batteries will be recharged. Figure 6.5shows the battery energy before and after the total excess energy is distributed. Therecharged energy for each battery will also be in the proportion of 1:2:3:4 (as the DOD),and therefore each battery receives 100 Wh, 200 Wh, 300 Wh, and 400 Wh, respectively.

PRIORITY EXCESS ENERGY SHARING

In this method, the excess energy is shared amongst the batteries in a ranked manner.The batteries are prioritized based on the DOD. Therefore, the highest DOD battery firstreceives the excess energy until it is full, then the next highest DOD-level battery gets the

6.2. METHODOLOGY

6

127

No share Prop. Pri. Eq.

1.2

1.4

1.6

1.8

2

2.2

Bat

tery

leve

l[kW

h]

SHS1 SHS2 SHS3 SHS4

Figure 6.5: Battery energy levels for the 4 SHS batteries without sharing and after the combined excess energydistribution based on proportional (Prop.) energy sharing, priority (Pri.) energy sharing, and equal (Eq.) energysharing. The rated battery capacity is also shown as the dashed line in the plot.

energy, if any, until it is full, and so on. In the context of the above example in Figure 6.5,the priority sharing would ensure that battery 4, owing to its highest DOD gets fullycharged to 2 kWh. The remaining 200 Wh goes to battery 3, which charges up to 1.6 kWh,same as battery 2 at the end of the time step.

EQUAL EXCESS ENERGY SHARING

In this method, the excess energy is shared equally amongst the batteries irrespective oftheir DOD. Therefore, in the context of the example shown in Figure 6.5, the equal sharingwould result in each battery getting 250 Wh of energy. It can be seen that battery 1 doesnot charge beyond 2 kWh, its maximum capacity, leading to a waste of 50 Wh. The equalsharing method can, therefore, be suboptimal in excess energy sharing.

It is interesting to note that these methods for sharing of the excess energy do notnecessarily create a uniform SOC across the microgrid. That is, SOC balancing of SHSbatteries in the microgrid is not the aim of these excess energy sharing methods.

CASE STUDY: HOMOGENEOUS MICROGRIDS

Two different case studies have been considered, viz., a homogeneous microgrid of tier4 SHS and a homogeneous microgrid of tier 5 SHS. Having a homogeneous microgridmakes for easier comparison in terms of the system metric gains per household. However,the study can also be extended to heterogeneous microgrids consisting of SHS belongingto different tiers and sizes.

Tier 4 microgrid Based on the optimal SHS sizes mentioned in Table 5.2, each SHSwas modelled with a PV size of 1740 Wp and battery size of 3560 Wh. Additionally, themicrogrid size was varied between 2 to 50 households. The averaged load profile perhousehold for the homogeneous tier 4 microgrid of 50 households is shown in Figure 6.6

6

128 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

along with the load profile for a single household for a representative day in the one-yearlong simulation.

0 4 8 12 16 20 240

200

400

600

800

1,000

Time [h]

Pow

er[W

]

SHSMicrogrid

Figure 6.6: Comparison of 1-day load profile for a tier 4 standalone SHS and an averaged load profile perhousehold for the SHS-based microgrid with 50 households for the same day. The demand diversity can be seento significantly reduce the peak average load.

Tier 5 microgrid Similar to the tier 4 case, each SHS was modelled based on the optimalSHS sizes (mentioned in Table 5.2), viz., a PV size of 4000 Wp and battery size of 6600 Wh.Additionally, the microgrid size was varied between 2 to 50 households. The averagedload profile per household for the tier 5 microgrid for 50 households is shown along withthe load profile for a standalone SHS for tier 5 in Figure 6.7.

SCOPE OF THE SHS-BASED MICROGRID STUDY

It should be noted that the concept of the interconnected SHS-based microgrid is exploredfrom the point-of-view of the overall benefits of energy sharing between SHS and themaximum potential consequent improvement in the system metrics for the microgrid asa whole. Accordingly, the study described in this chapter assumes perfect knowledge ofthe battery states of every SHS for sharing energy.

6.3. RESULTS AND DISCUSSIONBased on the modeling methodology described in Section 6.2.5, two SHS-based homo-geneous microgrids were modeled comprising tier 4 and tier 5 households respectively.Firstly, the concept of energy sharing in the microgrid as explained in Section 6.2.5 isdemonstrated in a scaled down case of 2 SHS. Secondly, using the optimal SHS sizes ofPV and battery for tier 4 and 5 as shown in Table 5.2, the microgrid was simulated for anincreasing number of households. The performance of the overall system in terms of LLPand Efail per household for increasing mcirogrid sizes is analyzed in Section 6.3.3. Finally,

6.3. RESULTS AND DISCUSSION

6

129

0 4 8 12 16 20 240

500

1,000

1,500

2,000

2,500

Time [h]

Pow

er[W

]

SHSMicrogrid

Figure 6.7: Comparison of 1-day load profile for a tier 5 standalone SHS and an averaged load profile perhousehold for the SHS-based microgrid with 50 households for the same day. The demand diversity can be seento significantly reduce the peak average load.

the benefits in the system sizing for the individual SHS participating in the microgrid caseis also shown in Section 6.3.4.

6.3.1. ENERGY EXCHANGE IN THE SHS-BASED MICROGRIDTo study the energy exchange behaviour, a reduced size of 2 SHS is initially considered inthe microgrid. Figures 6.8–6.10 depict the various parameters of the two SHS participatingin the microgrid over a period of 32 hours during the 1 year of simulation. It can be seenin Figure 6.8 that there is a relatively low PV production in the region around 6 hours,and higher PV production in the regions around 10-14 hours, and again around 30 hours.Consequently, as seen in Figure 6.9, both the SHS batteries are fully depleted at hour 6.Conversely, both the batteries are fully charged after the excess PV production aroundhour 30. After the initial high PV production around hour 12, however, only Battery 1 isfully charged.

The most remarkable feature of the energy sharing can be seen (in Figure 6.10) fromthe fact that neither SHS shows a positive Efail while there is even 1 non-empty battery.Additionally, until both the batteries are fully charged, the total Edump value in the inter-connected system is 0. A proportional excess energy sharing scheme was considered tocharge the battery with excess energy in this case.

6.3.2. COMPARISON OF BATTERY CHARGING USING EXCESS ENERGYFigure 6.11 shows the comparison between the 3 modes of excess energy-based batterycharging (described in Section 6.2.5) in tier 5 homogeneous microgrid in terms of averageLLP per household. The performance is measured over an increasing microgrid size upto 10 households. As seen in the figure, the priority excess sharing method yields the

6

130 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

0 6 12 18 24 300

1,000

2,000

3,000

Time [h]

Pow

er[W

]

Load1

Load2

PV

Figure 6.8: Load demand for each SHS and the PV generation per SHS at every time step.

0 6 12 18 24 300.2

0.4

0.6

0.8

1

Time [h]

SOC

[-] Batt1

Batt2

Figure 6.9: Battery SOC for each SHS at every time step.

0 6 12 18 24 300

20

40

60

80

Time [h]

En

ergy

[Wh

]

Efail,1

Efail,2

Edump

Figure 6.10: Efail for the 2 SHS and total Edump at every time step.

worst performing results. Proportional excess sharing method and equal sharing methodperform almost similarly for most microgrid sizes. Given the relative performance, forthe rest of the results, only the proportional sharing method for sharing excess energyis considered for the rest of the analysis while modeling the energy exchange in themicrogrid.

6.3.3. IMPACT OF MICROGRID SIZEFigure 6.12 shows the LLP per household for tier 4 and tier 5 homogeneous microgridsfor different microgrid sizes, up to a maximum of 50 houses. Each SHS in tier 4 has a PVand battery size of 1740 Wp and 3560 Wh respectively, while each tier 5 SHS has a PV andbattery size of 4 kWp and 6600 Wh respectively.

It can be seen that the gains in the average LLP per household due to the microgridlargely “saturate” after around 15 households for tier 4, and around 20 households for

6.3. RESULTS AND DISCUSSION

6

131

0 2 4 6 8 10

1.8

2

2.2

2.4

2.6

2.8

3·10−2

Number of households [-]

LL

P[-

]

Prop.Pri.Eq.

Figure 6.11: Performance comparison of the 3 excess energy-based battery charging methods in terms of LLPfor varying homogeneous (tier 5) microgrid sizes. The 3 methods are proportional (Prop.), priority (Pri.), andequal (Eq.) energy sharing

0 10 20 30 40 50

1

1.5

2

2.5

3

·10−2

Number of households [-]

LL

P[-

]

Tier 4Tier 5

Figure 6.12: Impact of increasing microgrid size on LLP per household for homogeneous microgrids of tiers 4and 5.

tier 5, respectively. This is because the load profiles per household for a given tier arequite similar, with the load profile peaks still reasonably close to each other. Therefore,the gains in these system metrics will significantly increase if there is a larger intra-day

6

132 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

variability between the load profiles for the various households. Additionally, it can beseen that the average LLP for tier 5 falls below 2%, which was not possible with purelystandalone SHS, wherein the minimum achievable LLP was 2.9% (Table 5.2).

0 10 20 30 40 50

0.8

0.85

0.9

0.95

1

Number of households [-]

No

rmal

ized

Efa

il[-

]

Tier 4Tier 5

Figure 6.13: Variations in normalized average Efail per household for homogeneous microgrids composed oftiers 4 and 5.

The average Efail per household also experiences a considerable decline with increas-ing size of the microgrid, as shown in Figure 6.13, where the normalized average Efail isrepresented for easy comparison between the tier 4 and tier 5 microgrids. The case witha single household is treated as the base value for normalization. The gains in averageEfail per household in the microgrid seems to largely stagnate after around 10 householdsfor both tier 4 and tier 5. While the tier 5 microgrid is able to achieve a reduction of 20%in the normalized average Efail per household, the tier 4 microgrid is able to achieve areduction of 10%.

It should be noted that this does not mean that the microgrid size should not grow tomore than 10 households. The fluctuations in these system metrics with every increasingmicrogrid size is attributed to the individual variability that an additional household’sload profile brings to the overall microgrid load profile. Figure 6.14 shows the impact ofincreasing microgrid size on the peak average load profile for the set of tier 5 load profilesconsidered in this study.

The peak average load largely decreases with increasing microgrid size, up to around10 households, beyond which there is much less reduction. Moreover, each additionalload profile brings about a tiny fluctuation in the peak average load, depending upon howcoinciding the individual load profile peak is with that of the overall microgrid. Conse-quently, greater variability in the load demand across the SHS will lead to greater energyexchange and therefore better performance in terms of the system metrics. Naturally,the use of any demand-side management technique resulting in a higher load profilevariability would result in more significant gains in each of the system metrics.

6.3. RESULTS AND DISCUSSION

6

133

0 10 20 30 40 50

1,800

2,000

2,200

Number of households [-]

Peak

aver

age

load

[W]

Figure 6.14: Change in peak average load with increasing microgrid size for tier 5.

In general, it can be said that an increase in microgrid size will improve the systemmetrics until a certain size, beyond which the gains in these metrics will largely stagnate.This directly follows from the fact that the variations in the total load profile for anycluster increase with cluster size until a certain point, beyond which the incrementalvariations in the cluster load profile are minimal. As the performance gains in the systemmetrics are limited by the load profile variability, these gains are expected to improvewhen considering heterogeneous microgrids, i.e., microgrids comprising householdsbelonging to mixed tiers of the MTF.

6.3.4. BENEFITS OF MICROGRID ON SHS SIZING

For the same level of performance on system metrics like LLP, an SHS-based microgridcan offer a significant advantage in the individual SHS sizing compared to a standaloneSHS. This is evident from Figure 6.15 as shown below, where for a particular microgrid size(number of households), the average LLP of the microgrid is evaluated across differentSHS battery sizes.

A microgrid size of 20 households has been considered in Figure 6.16 as that size wasseen to be enough (Figure 6.12) for average LLP of both tiers 4 and 5 to stabilize. The x andy-axes refer to the battery sizes and average LLP values respectively in the homogeneoustier 5 microgrid. Three different PV sizes are used corresponding to the optimal PV sizesobtained in Table 5.2 for each of the LLP thresholds.

It can be seen that the general nature of the LLP curve depending on the battery sizeremains the same as that of an individual standalone SHS as seen in other studies [3].The main difference is that with the energy sharing capability in the microgrid, the samesystem sizing would enable better performance. This can be seen while comparing the3 circular brown marks corresponding to the standalone system sizes from Table 5.2,with the 3 rectangular green marks corresponding to the same LLP threshold and PV size

6

134 6. QUANTIFYING THE BENEFITS OF SHS-BASED DC MICROGRIDS

3000 3500 4000 4500 5000 5500 6000 6500

0.02

0.04

0.06

0.08

0.10

Battery size [Wh]

LL

P[-

]

4050 Wp4000 Wp3800 Wp

Figure 6.15: Variations in average LLP per household over battery size for a homogeneous tier 5 microgrid with20 households. The points marked correspond to the different battery sizes for the standalone (circular dots)and microgrid case (green rectangles), respectively.

but lower battery sizes for the tier 5 microgrid case. More than 2 kWh of battery sizecan be saved to meet the same LLP threshold. In other words, battery sizing gains of41% (i.e., a 41% lower battery size) can be achieved for tier 5 SHS by energy sharing withthe microgrid as opposed to a standalone operation to meet LLP = 0.1. The gains arelower but still significant at 19.7% to meet LLP of 0.029, which was the minimum LLPachievable by an optimally sized tier 5 standalone SHS (Table 5.2).

A similar analysis is performed on a homogeneous tier 4 microgrid with 20 householdsto quantify the benefits achieved in terms of SHS battery sizes. This is shown in Figure 6.16.Similar to the tier 5 case, for all the 3 LLP thresholds shown in Table 5.2, significant batterysizing gains can be achieved through the microgrid when compared to standalone SHS tomeet the same LLP requirements. At 1740 Wh (blue rectangular mark in Figure 6.16), a31% lower sized battery is needed to meet an LLP threshold of 0.1 for a tier 4 microgrid, asopposed to a standalone SHS with a battery size of 2520 Wh (Table 5.2). Thus, the benefitsof energy sharing in an SHS-based microgrid have been quantified in this study.

Most importantly, as the microgrid proposed in this study is SHS-based, the mainproblems of centralized PV battery installations with substantial component costs canbe avoided. Additionally, each SHS belongs to a different household participating in themicrogrid, as opposed to centralized microgrids with lack of clear ownership.

6.4. CONCLUSIONSThis study presented a detailed methodology for modelling an interconnected SHS-basedmicrogrid. A modularly expandable and scalable microgrid architecture is proposed.Moreover, meeting the energy demand of higher tiers like tier 4 and tier 5 is shown tobe possible with this approach of bottom-up, interconnected SHS-based microgrid as

REFERENCES

6

135

2000 2500 3000 3500 4000

0.02

0.04

0.06

0.08

0.10

Battery size [Wh]

LL

P[-

]

1740 Wp1620 Wp1500 Wp

Figure 6.16: Variations in average LLP per household over battery size for a homogeneous tier 4 microgrid with20 households. The points marked correspond to the different battery sizes for the standalone (circular orangedots) and microgrid case (blue rectangles), respectively.

opposed to standalone SHS. The energy sharing between the different SHS is modeled andits positive impact on the system metrics quantified. It is shown that battery sizing gainsof more than 40% can be achieved at the tier 5 level with interconnectivity as compared tostandalone SHS to meet the same loss of load probability threshold. It should be noted thatthe sizing gains and better performance with respect to the system metrics can increaseeven further depending on the demand diversity in the load profiles. Furthermore, such abottom-up microgrid can also help in utilizing high power appliances, especially the onesthat cater to productive use of energy.

RECOMMENDATIONS AND FUTURE WORKThe interconnected SHS-based microgrid described in this chapter assumes perfectknowledge of the dynamic behaviour of every SHS. For a truly decentralized bottom-upexpansion of such an envisioned microgrid, decentralized power management shouldbe considered. Additionally, optimal microgrid topologies should be studied for sucha decentralized microgrid. Moreover, the effects of load profile variations within themicrogrid along with load and PV generation uncertainty can also be examined.

REFERENCES[1] Lighting Global and Dalberg Advisors, “Off-grid solar market trends report,” tech.

rep., Lighting Global, Dalberg Advisors, GOGLA, ESMAP, 2018.

[2] N. Narayan, J. Popovic, J. C. Diehl, S. Silvester, P. Bauer, and M. Zeman, “Developingfor developing nations: Exploring an affordable solar home system design,” in 2016IEEE Global Humanitarian Technology Conference (GHTC), pp. 474–480, Oct 2016.

6

136 REFERENCES

[3] N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Exploring the boundaries of solar home systems (shs) for off-grid elec-trification: Optimal shs sizing for the multi-tier framework for household electricityaccess,” Applied Energy, vol. 240, pp. 907 – 917, 2019.

[4] F. A. Bhuiyan, A. Yazdani, and S. L. Primak, “Optimal sizing approach for islandedmicrogrids,” IET Renewable Power Generation, vol. 9, no. 2, pp. 166–175, 2015.

[5] F. Benavente-Araoz, A. Lundblad, P. E. Campana, Y. Zhang, S. Cabrera, and G. Lind-bergh, “Loss-of-load probability analysis for optimization of small off-grid pv-batterysystems in bolivia,” Energy Procedia, vol. 142, pp. 3715 – 3720, 2017. Proceedings ofthe 9th International Conference on Applied Energy.

[6] A. Smets, K. Jäger, O. Isabella, R. Van Swaaij, and M. Zeman, Solar Energy - The physicsand engineering of photovoltaic conversion, technologies and systems. 02 2016.

[7] H. Kirchhoff, “Identifying hidden resources in solar home systems as the basis forbottom-up grids,” in Decentralized Solutions for Developing Economies, pp. 23–32,Springer, 2015.

[8] B. Soltowski, J. Bowes, S. Strachan, and O. limpo Anaya-Lara, “A simulation-basedevaluation of the benefits and barriers to interconnected solar home systems in eastafrica,” in 2018 IEEE PES/IAS PowerAfrica, pp. 491–496, IEEE, 2018.

[9] A. Chaurey and T. Kandpal, “A techno-economic comparison of rural electrifica-tion based on solar home systems and pv microgrids,” Energy Policy, vol. 38, no. 6,pp. 3118 – 3129, 2010. The Role of Trust in Managing Uncertainties in the Transitionto a Sustainable Energy Economy, Special Section with Regular Papers.

[10] A. Bani-Ahmed, M. Rashidi, and A. Nasiri, “Coordinated failure response and re-covery in a decentralized microgrid architecture,” in 2017 IEEE Energy ConversionCongress and Exposition (ECCE), pp. 4821–4825, Oct 2017.

[11] Y. Xia, W. Wei, Y. Peng, P. Yang, and M. Yu, “Decentralized coordination control forparallel bidirectional power converters in a grid-connected dc microgrid,” IEEETransactions on Smart Grid, vol. 9, pp. 6850–6861, Nov 2018.

[12] Z. Wang, F. Liu, Y. Chen, S. H. Low, and S. Mei, “Unified distributed control of stand-alone dc microgrids,” IEEE Transactions on Smart Grid, vol. 10, pp. 1013–1024, Jan2019.

[13] Z. Liu, M. Su, Y. Sun, L. Li, H. Han, X. Zhang, and M. Zheng, “Optimal criterion andglobal/sub-optimal control schemes of decentralized economical dispatch for acmicrogrid,” International Journal of Electrical Power and Energy Systems, vol. 104,pp. 38 – 42, 2019.

[14] M. Baranwal, A. Askarian, S. Salapaka, and M. Salapaka, “A distributed architecturefor robust and optimal control of dc microgrids,” IEEE Transactions on IndustrialElectronics, vol. 66, pp. 3082–3092, April 2019.

REFERENCES

6

137

[15] M. S. Rahman, M. J. Hossain, J. Lu, and H. R. Pota, “A need-based distributed co-ordination strategy for ev storages in a commercial hybrid ac/dc microgrid withan improved interlinking converter control topology,” IEEE Transactions on EnergyConversion, vol. 33, pp. 1372–1383, Sep. 2018.

[16] H. R. Baghaee, M. Mirsalim, G. B. Gharehpetian, and H. A. Talebi, “Decentralizedsliding mode control of wg/pv/fc microgrids under unbalanced and nonlinear loadconditions for on- and off-grid modes,” IEEE Systems Journal, vol. 12, pp. 3108–3119,Dec 2018.

[17] K. Gong, E. Lenz, and U. Konigorski, “Decentralized frequency control of a ddg-pvmicrogrid in islanded mode,” in 2015 23rd Mediterranean Conference on Control andAutomation (MED), pp. 292–297, June 2015.

[18] “Rural Spark - Energy Kit.” http://www.ruralspark.com/products/energy-kit/. (Date last accessed 2015-10-12).

[19] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrifica-tion,” in 2017 International Conference on the Domestic Use of Energy (DUE), pp. 8–15,April 2017.

[20] N. Narayan, Z. Qin, J. Popovic-Gerber, J.-C. Diehl, P. Bauer, and M. Zeman, “Stochasticload profile construction for the multi-tier framework for household electricityaccess using off-grid dc appliances,” Energy Efficiency, Nov 2018.

[21] B. Soltowski, S. Strachan, O. Anaya-Lara, D. Frame, and M. Dolan, “Using smartpower management control to maximize energy utilization and reliability withina microgrid of interconnected solar home systems,” in 7th Annual IEEE GlobalHumanitarian Technologies Conference, GHTC 2017, pp. 1–7, 2017.

[22] S. Groh, D. Philipp, B. E. Lasch, and H. Kirchhoff, “Swarm electrification: Investigat-ing a paradigm shift through the building of microgrids bottom-up,” in DecentralizedSolutions for Developing Economies, pp. 3–22, Springer, 2015.

[23] S. Groh, D. Philipp, B. E. Lasch, and H. Kirchhoff, “Swarm electrification-suggestinga paradigm change through building microgrids bottom-up,” in Developments inRenewable Energy Technology (ICDRET), 2014 3rd International Conference on the,pp. 1–2, IEEE, 2014.

[24] Meteotest, “(software) meteonorm ver 7.1,” 2014.

[25] N. Narayan, “Electrical power consumption load profiles for households with dcappliances related to multi-tier framework of household electricity access.” https://doi.org/10.4121/uuid:c8efa325-87fe-4125-961e-9f2684cd2086, 2018.

[26] N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Zeman, “Amodeling methodology to evaluate the impact of temperature on solar home systems

6

138 REFERENCES

for rural electrification,” in 2018 IEEE International Energy Conference (ENERGY-CON), pp. 1–6, June 2018.

[27] A. Jhunjhunwala, A. Lolla, and P. Kaur, “Solar-dc microgrid for indian homes: Atransforming power scenario,” IEEE Electrification Magazine, vol. 4, no. 2, pp. 10–19,2016.

[28] N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Estimating battery lifetimes in solar home system design using a practi-cal modelling methodology,” Applied Energy, vol. 228, pp. 1629 – 1639, 2018.

7OPTIMAL MICROGRID LAYOUT

USING GEOGRAPHIC INFORMATION

SYSTEM (GIS)

The saddest aspect of life right now is that science gathers knowledge fasterthan society gathers wisdom

Isaac Asimov

ABSTRACTAs discussed in Chapter 6, bottom-up meshed DC microgrids born out of the interconnec-tion of SHS are instrumental to achieving higher tiers of electrification while maintainingscalability. However, there is limited knowledge when considering the possible layoutsfor such microgrids, as conventional microgrid topology (layout) studies are limited tocertain standard layouts, like spider, ring, bus, and radial layouts. The optimal micro-grid layouts in the context of (remote) rural electrification can be quite different fromthe conventional layouts, because of various reasons like geographical spread, lack ofstructured housing layouts, etc., thereby necessitating a different methodology altogetherwhen thinking of microgrid planning for rural electrification. In this chapter, a geographicinformation system (GIS)-based methodology is introduced, which takes into account theactual geographic spread of households in remote areas. After GIS-based data processing,graph theory concepts are used to minimize the layout costs. A total of 42 different remotesites around the world are considered in the study. The proposed integrated methodologyis explained in this chapter with an example of its application to a sample of the remoteareas from the 42 different sites. The results of the layout comparisons show how theconventional microgrid layouts can be easily outperformed in terms of network length

This chapter is based on: N. Narayan, M. Tagliapietra, Z. Qin, J. Popovic-Gerber, P. Bauer, & M. Zeman. Optimalmicrogrid layout using Geographic Information System and graph theory concepts, submitted.

139

7

140 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

by the microgrid layouts arrived at through the integrated approach proposed in themethodology, highlighting the usefulness of the approach.

OUTLINE

This chapter is divided into five sections. Section 7.1 introduces the work describedin the chapter with a literature review on the other studies that have used geographicinformation systems (GIS) for planning electricity networks. Section 7.2 outlines thenecessary technical background in the different related fields — GIS and graph theoryapplied to network analysis — while also defining the focus of this study. Section 7.3describes the methodology used to arrive at optimal microgrid layouts. Section 7.4discusses the results and compares the performance of different microgrid layouts acrosspre-defined metrics. Finally, Section 7.5 concludes the discussion on the GIS-basedmethodologies to determine optimal microgrid layouts.

7.1. INTRODUCTIONOne of the decisions that need to be taken in microgrid planning based on SHS is thechoice of the microgrid layout. However, there is limited knowledge when considering thepossible layouts (topologies) for such microgrids, as conventional microgrid topologiesare limited to certain standard layouts, like spider, ring, bus, and radial layouts. On theother hand, the real-world spatial spread of the households in a given area might deem aparticular standard layout more expensive due to higher cabling costs. It could also bethat a different layout altogether might be needed for a specific case. In this chapter, anintegrated geographic information system (GIS)-based methodology is proposed thattakes into account the ground-level data, uses graph theory concepts to reduce cablingcosts, and finally uses an optimization algorithm to strike a balance between cables costsand the operational parameters of the microgrid like line losses and line congestion,suitable to be applied at a pre-electrical design phase of the microgrid planning.

7.1.1. LITERATURE REVIEW

CONVENTIONAL MICROGRID TOPOLOGY STUDIES

With regards to the conventional microgrid topology studies, the range of topologiesexplored in the past is limited. Before the diffusion of renewables, the easiest way toprovide energy to remote communities was to install centralised generators (such asdiesel gensets) to which all households were connected via two possible layouts. First wasthe spider topology or spider diagram, sometimes also called as the hub-and-spoke layout[1], meaning that each household is directly connected to the generator by a dedicatedcable segment. Second was the radial topology, sometimes also called the hub-and-trunklayout [1], in which multiple branches span from the central generator, reaching eachhousehold in a consecutive way along the branch, with the possibility of sub-branchesbeing implemented [2, 3]. These topologies present a major drawback in terms of faulttolerance, since if one of the cable segments is disconnected because of a failure, thehouseholds which are connected to that branch will be totally cut out from energy supplyand will suffer major power outages, especially in the case of centralized microgrids [4].

In the context of decentralized SHS-based microgrids, however, the fault tolerance of

7.1. INTRODUCTION

7

141

(a) Spider topology (b) Radial topology

Figure 7.1: Examples of radial topology and spider diagram adapted to the context of decentralized microgridwith SHS.

these layouts can be inherently better than centralized microgrids. Nonetheless, energysharing can still be impacted by failures. A conceptual illustration for the radial and spidertopologies adapted to SHS-based decentralized microgrids can be seen in Figure 7.1. Eventhough those layouts are typical of centralized generators, they are going to be anywayincluded into the comparative analysis between decentralized microgrid layouts in thisstudy, as seen later in Section 7.3.

Another kind of topology, which is actually applicable both to centralised and decen-tralised microgrids, is called the bus topology, which consists of a single (DC) bus to whichall households are singularly connected (Figure 7.2(b)). In [2], a comparative analysis ofthis DC bus layout and the already mentioned centralised radial topology is carried out,highlighting how the distributed bus solution is more easily modularised and scalable,fitting in the framework of a bottom-up approach to electrification.

(a) Ring topology (b) Bus topology

Figure 7.2: Examples of ring and bus topologies adapted to the context of decentralized microgrid with SHS.

Another possible layout option, which is mentioned and described, among others,in [3] and [5], is called the ring topology, consisting of a consecutive connection of allthe involved households, forming a closed loop which spans through and around thevillage/neighbourhood, as depicted in Figure 7.2(a).

The last kind of topology is the meshed grid topology. Meshed grids have the peculiarcharacteristics of having redundant cable segments, providing alternative paths for energy

7

142 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

exchange. One major advantage of meshed microgrids is an highly increased fault toler-ance, since even if one or multiple cable segments fail or are tampered, the alternativepaths manage to keep the whole microgrid functioning properly. Another advantage isgiven by a more even voltage level distribution, which leads to lower fluctuations anddifferences in voltage between different sections of the grid [3, 6]. The obvious downsideof this kind of topology is a more complex layout and higher costs due to more cablelength needed to connect households with redundant paths.

GIS-BASED STUDIES

There have been a limited number of GIS-based studies in the past related to electrifica-tion planning. However, they differ in scope and considerations with respect to the studydescribed in this chapter, as explained in this section.

Most GIS-based studies for electrification planning found in literature are focused onspecific geographical areas. For example, a GIS-based decision support tool is discussedspecifically for northeast of Brazil in [7], a cost-effective electrification is sought for Kenyabased on GIS in [8], a GIS-based Network Planner tool is presented in [9] to compare theelectrification options in Ghana, another Network Planner tool-based study is presentedfor the case of Nigeria in [10], and site selection for micro-hydro power generation systemsin the eastern Himalayan region of India is enhanced using GIS in [11]. GIS is also usedto perform a cost comparison of technology approaches to improve energy access withcase studies for Ethiopia and Nigeria in [12], a fully GIS-based energy access modelingis performed for a case study for Ethiopia in [13], another case study on Ethiopia isconsidered while using geospatial energy access planning in [14], and another GIS-basedcase-study on Nigeria for electrification planning is detailed in [15].

In terms of applications, the GIS-based studies reported in literature so far encom-pass a range of applications within the context of electrification planning. For instance,performance of PV mini-grid systems over large geographical areas is determined usingGIS in [16]. In [9], data on electricity demands and costs is combined with populationand other socio-economic data while computing demand estimates, with the aim foridentifying the most viable electrification alternative amongst grid extension, standaloneSHS, and mini-grid. Additionally, the microgrids considered are AC and also diesel-based.Again, least cost electrification is sought in [8] between standalone SHS and grid extensionwithout considering microgrids. Similarly, a least cost model is built in [12] to comparetechnology costs of electrification through multiple renewable technology-based mini-grids, standalone PV, and grid extension. However, diesel is also considered as one of thesupply technologies for both standalone systems and mini-grids. Population density isconsidered along with costs of grid-based electricity for the case-studies in Nigeria andEthiopia. The concept of multitier framework (MTF) for measuring energy access is alsoconsidered, while the two cost-based parameters considered in [12] are levelized cost ofelectricity (LCOE) and total costs per household. Both urban and rural areas of Nigeria areconsidered in the study described in [15], which uses GIS-based data to arrive at LCOE fordifferent electrification options including diesel gensets. A similar analysis is undertakenin [14] for Ethiopia.

A variety of energy technologies is considered including PV, biomass, wind, diesel andhydro in [17], which develops an Open Source Spatial Electrification Toolkit (OnSSET)

7.1. INTRODUCTION

7

143

using GIS and an integrated electrification model. Different considerations like householddensity, local renewables and diesel costs, local grid distances, and kWh/household/yearfor various tiers of the MTF are taken into account in [17], in sub-Saharan Africa. In [7],renewable energy management is the primary goal while using GIS for the semi-aridenvironments in north-eastern Brazil. Biomass is also considered as an energy sourcein [7]. Satellite imagery is used in [18] for the purpose of identifying extremely poorvillages suitable for unconditional cash transfers for poverty alleviation and siting PV ruralmicrogrids. GIS is used for the purpose of creating energy access projections in [13]. Arecent study has also integrated urban energy models within GIS into an open sourceplatform for the optimization of energy systems in cities, which also helps in the optimalallocation of distributed storage in urban energy systems [19].

As seen in this section, while GIS has been used for electrification planning in general,there is a clear research gap in terms of investigating microgrid topologies or layouts usingGIS. Furthermore, the microgrids considered in this chapter are decentralized, basedon SHS, and are purely DC in nature. Additionally, graph theory-based concepts andspatial geometry are utilized in this study to arrive at the optimal microgrid layouts. Inthis chapter, using GIS as the first step, a methodology is developed to identify the optimalmicrogrid layouts for the purpose of off-grid electrification.

7.1.2. SCOPE OF THIS STUDY

The work described in this chapter is limited to a pre-electrical design phase study, whichcan already give insights towards the optimal topology for a decentralized microgrid. Themain aim of this work is the generation, analysis and comparison of different microgridtopologies at the scale of an intra-village or community interconnection, using graphtheory concepts for gaining insights towards cost considerations as well as some electricaloperational parameters like line congestion and redundancy, starting from real ground-level data.

7.1.3. CONTRIBUTIONS OF THIS CHAPTER

Following are the main contributions of this chapter.

1. Incorporation of ground-level data through GIS. The use of GIS enables the effi-cient analysis of high resolution ground level geographic data, which reflects theactual physical spread of the households in remote areas.

2. Graph theory-based analysis. A graph theory-based analysis at a village or com-munity level offers a more realistic methodology than depending on conventionalmicrogrid layouts that may not be tailored for the specific geographic spread of thehouseholds at a given site.

3. Passive optimization at the design stage. The methodology proposed in this chap-ter enables a passive optimization at the design stage between the cabling costsand the operational parameters like line congestion without necessitating an activepower flow analysis.

7

144 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

7.2. GRAPH THEORYGraph theory is a branch of mathematics that deals with the study of mathematicalelements called graphs. This section outlines a few basic concepts from graph theory anddescribes how they can be used in the context of evaluating microgrid topologies.

7.2.1. GRAPH THEORY: SOME CONCEPTS AND DEFINITIONS

GRAPH

A graph G = (V ,E) is a pair of subsets, in which V is a set of n vertices (or nodes) and Eis a set of m edges (lines). Each edge e in the set E is an object that connects betweenthem two nodes i and j of the set V . An edge is said to be incident to the two vertices itconnects, while each of the two vertices can be said to be adjacent to each other whenconnected through a common edge.

Edges can also be created by connecting a single vertex to itself; in this case the edgeis said to be a self-loop. However, given that this study deals with the use of graph theoryonly in the context of electricity networks and microgrids, such self-loops will not beconsidered.

EDGE WEIGHTS

Edges are usually assigned a weight, an additional parameter for the analysis of thenetwork characteristics. The weight can be any kind of network characteristic dependingon the kind of network being considered and the nature and purpose of the analysis. Forexample, length of each edge in meters can be used a weight for the specific purpose ofmicrogrid topology analysis.

(UN)DIRECTED GRAPHS

An undirected graph is one where each edge is only defined by an unordered pair ofvertices (i , j ) it connects, along with its weight. On the other hand, a directed graph, alsosometimes called a Digraph, is a graph with oriented edges. That is, an edge now has tobe additionally defined by its direction. Therefore, in the case of digraphs, the orderedcouple of vertices (i , j ) will be different from ( j , i ). Figure 7.3 depicts simple examples ofdirected and undirected graphs.

(a) Undirected graph with 5 verticesand 4 edges.

(b) Directed graph with 5 vertices and4 directed edges

Figure 7.3: Examples of directed and undirected graphs.

7.2. GRAPH THEORY

7

145

DEGREE OF CONNECTIVITY

Degree of connectivity of a vertex is defined as the number of edges incident on the vertex.Consequently, a few special vertices can be classified based on their degree of connectivity.For example, isolated vertices have a degree of 0 as they are completely disconnectedfrom the rest of the graph; vertices with a degree of connectivity 1 are called end verticesas they are the dead-end branches in a graph; vertices in a graph with n edges and havinga degree of connectivity of n −1, i.e., connected to every other vertex through a singleedge, are called dominant or universal vertices. An example of a dominant vertex is thecentral node of the spider diagram considered in Figure 7.1(a).

PATHS AND (UN)CONNECTED GRAPHS

A path is a sequence of edges connecting an ordered sequence of vertices in a graph. Apath from vertex i to vertex j is a sequence of edges starting from vertex i and finishing atvertex j going through a finite or infinite number of vertices along the way. Consequently,an edge (i , j ) in E can be a possible path from i to j , but need not be the only path.

A connected graph is one in which there is at least one existing path connectingany pair of vertices i , j in V through any path or a number of edges in E . An exampleillustration can be seen in Figure 7.4(c). On the other hand, if even one pair of verticesin V remains unconnected, then the graph is said to be unconnected. By definition, anunconnected graph is devoid of any universal vertices. An example of unconnected graphcan be seen in Figure 7.4(b).

Additionally, the concepts of empty and complete graphs can be introduced. A graphG = (V ,E ) is said to be an empty graph if the set E is empty. In other words, no edges existin the graph. Alternatively, it can be said that the degree of connectivity is 0 for all thevertices in V . Figure 7.4(a) illustrates an empty graph. Finally, a complete graph is onein which each vertex i in V is connected to every other vertex j through a direct edgein E . Therefore, a complete graph is composed of only universal vertices. Figure 7.4(d)illustrates a complete graph. Furthermore, it can be said that the number of edges m in acomplete graph is given by Equation 7.1.

m = n(n −1) for a digraph (7.1)

= n(n −1)

2for an undirected graph

(a) Empty (b) Unconnected (c) Connected (d) Complete

Figure 7.4: Types of graphs for the same set of vertices V .

Given that this study focuses on microgrid topologies, the work described here will

7

146 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

focus on connected graphs. In other words, the microgrid is assumed to connect allhouseholds in the given settlement.

MATRIX REPRESENTATION

For using graphs in the context of computation, two particular matrix structures can helpin representing correlation between nodes or vertices and edges, viz., adjacency matrixand incidence matrix.

An adjacency matrix A is an n×n matrix, representing a graph G in terms of adjacencybetween its vertices, as shown in Equation 7.2.

Ai j = 0∀(i , j ) ∉ E i.e., if i and j are not adjacent (7.2)

Ai j = 1∀(i , j ) ∈ E i.e., if i and j are adjacent

Consequently, it can be said that the adjacency matrix for an empty graph will haveAi j = 0∀i , j , while the adjacency matrix for a complete graph will have Ai j = 1∀i 6= j , asthe diagonal elements being equal to 1 would represent self-loops.

An incidence matrix I is an m ×n matrix, representing a graph G in terms of theincidence of each edge e in E with each vertex v in V , as shown in Equation 7.3 for anundirected graph.

Ive = 0 i.e., if e is not incident with v (7.3)

Ive = 1 i.e., if e is incident with v

The definition of each edge in the matrix is therefore dependent on the two verticesit connects. For a directed graph, an additional information on direction needs to bepresent. This is typically done by identifying the start vertex or node with a value of -1and end node with a value of 1 in each edge column e.

7.2.2. TREES IN GRAPH THEORYA tree in graph theory can be defined as an acyclic connected undirected graph, or anundirected graph without any loops. In other words, each pair of nodes is connectedby means of a single unique path. For any given set of n nodes, a tree is, by definition,composed of m = n −1 edges. A tree is a sub-graph of the given graph G = (V ,E), whichconnects at the same time all the vertices included in a given subset of V , using edgesincluded in E . If all the vertices in V are connected by a single tree, that tree is then calleda spanning tree. The concept of trees is very important in the area of network analysis,and is also used in this study. A thorough and exhaustive work on trees in graph theorycan be found in [20].

MINIMUM SPANNING TREE

The concept of trees can be applied to many practical problems in which the aim is toconnect a subset of elements (nodes, vertices, people, houses, etc. depending on the kindof network) using the lowest possible number of relations (edges, lines, connections, etc.depending on the kind of network). These kinds of problems usually have cost minimiza-tion as one of the main criteria for connecting the desired elements. Therefore, one of themain geometrical problems addressed in graph theory addressed by mathematicians is a

7.2. GRAPH THEORY

7

147

way to obtain the shortest or least costly spanning tree for a given subset of vertices in Vfor a graph G . Such a sub-graph or tree connecting all the vertices without any cycles witha minimum possible total edge-weight is called a minimum spanning tree (MST).

Figure 7.5: Example of a minimum spanning tree (MST) for a weighted graph. The MST as a sub-graph is shownin bold in the otherwise connected graph.

The problem of MST, by definition, is only applied to weighted edges. Figure 7.5shows an example of an MST for an undirected graph with weighted edges. The MSTproblem has been discussed in literature for decades, with many algorithms having beendeveloped to solve it already since the 1950s [21], the most known and efficient examplesbeing Kruskal’s and Prim’s algorithms.

Prim’s algorithm, which is also used later to generate the MST layout, utilizes theconcept of the nearest neighbour, which means the vertex which has the shortest distance(weight) in G from a given vertex or sub-tree. The idea behind Prim’s algorithm can beexplained as follows [22].

1. Take the starting graph G = (V ,S) for the set of vertices V , knowing the weight ofeach single edge contained in S.

2. Start creating a tree T = (V ,E), E being an empty set of edges.

3. Perform one of the following actions:

(a) Add to E the edge in S connecting any vertex in V to its nearest neighbour.

(b) Add to E the edge in S connecting any sub-tree in T to its nearest neighbour.

4. Repeat step 3 until a spanning tree is formed in T .

5. The obtained spanning tree T is an MST for the set of vertices V and the set of edgesS.

7

148 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

This algorithm is particularly useful for Euclidean applications to connect points ina geometrical plane using Euclidean distance as weight. This is because the weights ofthe potential edges do not necessarily need to be known aforehand as an input to thealgorithm as long as the geometrical locations of the points are known, which can be usedto calculate the distance between each pair of points.

7.2.3. GRAPH THEORY APPLIED TO NETWORK ANALYSIS

Myriads of real-world applications exist for graph theory. Some common examplesinclude planning (or maintenance) of road networks, water supply networks, electricitytransmission grid, GPS navigation systems, among many others. For the study describedin this chapter, graph theory is going to be used as a tool to evaluate, analyze and comparemicrogrid layouts or topologies. Therefore, the vertices of the graphs would be the singlenodes of the microgrid, which in this study are the single households inside a communitywith SHS that need to be interconnected. The edges of the graph would be the lines(cables) of the microgrid’s electrical distribution network that is physically connecting thehouseholds.

One major advantage of using these concepts is the great flexibility they offer, makingit viable to study different parameters and aspects of microgrids with just minor modifi-cations. As an example, different aspects of the microgrids can be analyzed consideringdifferent characteristics of the lines as weight of the edges. For example, using the physicaleuclidean length of the cables as weight is a good way to analyze and compare cable costsfor different layouts, while using cable resistances as weight can be useful for power flowanalysis and for losses calculations.

Examples of graph theory applied to electrical network problems are already foundin literature with different scopes and purposes. An extensive overview of how electricalnetworks can be analyzed and represented using graph theory, focusing mainly on thecircuit modelling point of view, is given in [23]. Graph theory application to networkplanning at the microgrid level has been suggested in literature, usually focusing on justone algorithm or one suggested starting topology, without usually making an extensivecomparison of different possible solutions [24].

AVERAGE NETWORK LENGTH

The most obvious parameter for the comparison and bench-marking of topologies is thenetwork length itself. The average network length in this context can then be definedas the distance that needs to be covered with cables per household for each topology.Naturally, for cost considerations, it is desirable to keep the average network length aslow as possible. It should be noted that the actual costs of installing a microgrid wouldinclude other aspects such as poles, interfacing converters, etc. However, for the purposeof comparing the microgrid layouts with each other, the average network length is deemedto be a sufficient first indicator to give insights on cost comparisons.

Given the least total edge length and therefore the least average network length, theMST will be considered as an additional layout apart from the 4 conventional topologiesshown in Figures 7.1 and 7.2 when considering the microgrid topologies in this study.

7.3. METHODOLOGY

7

149

AVERAGE SHORTEST PATH LENGTH

In general, the shortest path problem aims to find the path between two vertices i andj in a weighted graph G , which has the lowest possible sum of the weights of the edgesincluded in the path. Many different versions of the problem can be formulated, de-pending on the graph being directed or undirected and the vertices that are taken intoconsideration. The simplest version is the single-pair shortest path problem, which con-siders only one pair of vertices i and j , i being the starting point of the path and j theend point of the path. There are many algorithms providing solutions to the shortestpath problem, like the Dijkstra algorithm [25] and the Floyd-Warshall algorithm [26].The all-pairs shortest path length problem is addressed in these, which is particularlyuseful for the calculation of the average shortest path length of a graph or network bysimply averaging the length of all the shortest paths found by the algorithm. As explainedlater in Section 7.3, the average shortest path length can be used as a representationof some network characteristics, such as redundancy, congestion avoidance and lossreduction. Hence, this parameter is going to be used for microgrid topology comparisonand optimization.

7.3. METHODOLOGYIn order to arrive at the optimized microgrid topologies, a series of steps is carried out,starting with obtaining real-world geographical inputs from a GIS software.

7.3.1. GIS-BASED DATA PROCESSINGThe overall GIS-based methodology is depicted in the flowchart shown in Figure 7.6. Theoutput topologies obtained from this GIS-based methodology will then be used in thelayout optimization described in Section 7.3.3.

Identification oftarget villages

Creation of shape-files containing lo-cation information

Creation of differentnetwork topologies

Layout optimization

Figure 7.6: Block diagram explaining the overall methodology based on QGIS. The shaded block involving layoutoptimization is explained in detail in Section 7.3.3 and Figure 7.14.

QUANTUM GIS (QGIS)In general, a Geographic Information System can be defined as any system which has thepurpose to collect, analyze, process, represent and communicate spatial and geographic

the computation of the shortest paths from one single vertex i to all other vertices in the graph, for all pairsof vertices (i , j ) at once.

7

150 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

data. For the purpose of this study, a fully open source GIS software called QuantumGIS (QGIS) was used [27]. QGIS has an interface that provides a visualization windowfor geographical features, a customized toolbar, a browser panel, a layer panel and auseful built-in Python console, including an internal editor, as shown in Figure 7.7. All thedata is stored in layer files, with the format being shapefiles (.shp), which is a compositeensemble of different files storing geometrical features (points, lines or polygons), theirgeographical location, and a given set of attributes. The version of QGIS used in this studywas "QGIS 2.18 las Palmas de Gran Canaria".

Figure 7.7: Snapshot of the QGIS interface.

LOCATION SELECTION

The first step using QGIS was the identification of a sample of (remote) locations to beused as case studies, from different (rural) parts of the world. It was decided to look forremote areas in different countries and different environments, to give a varied sample,and to focus mainly on small villages, with a number of households up to around 40.In the process of choosing the smaller villages, it was also chosen to select settlementsshowing different spatial household distribution patterns for sake of variety and diversityof the sample. Part of the sample was selected starting from available databases forreal-world rural electrification projects, while other locations were chosen from specifictargeted countries with energy access issues and based on their distance from availableinfrastructure. For example, the set of Tanzanian villages were retrieved from the RuralElectrification Densification Programme (REDP) database [28]. Similarly, for villages inBangladesh, an available database on "hard-to-reach" communities was used to pickthese locations [29]. This resulted in the selection of 42 locations, spread over 12 different

7.3. METHODOLOGY

7

151

countries, as shown in Figure 7.8.

Figure 7.8: The selected locations for the case-studies shown on the world map.

The sample locations, as shown in Table 7.1, ranges from 5 to 40 households perlocation, from a density of around 100 households per km2 to around 5000 householdsper km2 and from a surface area of around 3000 m2 to around 150000 m2. The countrieswhich were included into the analysis are Tanzania, Gabon, Cameroon, Madagascar,Bangladesh, India, Nepal, Cambodia, Colombia, Peru, Brazil and Iceland. Iceland wasincluded just to show how the methodology can also be applied to isolated areas ofdeveloped nations. All the other selected countries at the time of conducting this studyhad low rural electrification rates and the chosen locations (villages) were disconnectedfrom the main grid.

Table 7.1: Ranges of parameters for the selected locations.

Min Max

Number of Households [hh] 5 40Density [hh/km2] 93.58 5016.03Area [m2] 2851 145324

With the villages identified, the next step was to create shapefiles containing informa-tion about the locations. Two types of shapefile layers were created:

• Point shapefiles, containing point features representing households of the locations

• Polygon shapefiles, containing the external perimeter of the locations

7

152 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

After having created the shapefiles of the locations, it was possible to start the analysisof the gathered information, using the geographical correlation based on coordinateproximity between the separate shapefiles. Figure 7.9 shows a sample of 3 of the locationswith identified households and their external perimeters.

Figure 7.9: Sample of the locations with identified households and external perimeter. From left to right:locations 10, 15, and 26.

Once the two shapefiles and the mentioned attributes had been created, the nextstep in the QGIS-based analysis was to create different microgrid layouts to connect thehouseholds between each other.

7.3.2. MICROGRID TOPOLOGY CREATIONFive different kind of microgrid topologies were identified and chosen to be comparedin this study, viz: 1. Spider diagram 2. Ring topology 3. Bus topology 4. Radial topology5. Minimum Spanning Tree (MST).

Before the creation of these topologies for each location, one additional feature hadto be identified, viz., the center of the village, which is the point having the minimumaverage distance from the households. The (x,y) coordinates of this central point werefound using Equation 7.4.

xc =∑hh

k=1 xk

hh(7.4)

yc =∑hh

k=1 yk

hh

in which hh is the number of households of the given village, xc and yc are thecoordinates of the center of the village and xk and yk are the coordinates of each of thehouseholds inside a given village. It must be noted that in this case the same relevance(weight) was assigned to each household. It is possible to multiply each coordinateby a relative weight if, for example, different household power production and energydemands need to be taken into account. These calculations to find the center of thelocation can be performed easily within QGIS.

SPIDER DIAGRAM

Starting from the spider diagram in which every single household is connected directlyto a central point (in this case the just identified center of the village), a tool from the

7.3. METHODOLOGY

7

153

QGIS repository called "RT QSpider" can be used. This tool creates a line shapefile layercontaining multiple line geometries, taking as inputs starting and ending (x,y) coordinatesfrom a defined input layer. In this case, the starting coordinates are the single householdcoordinates, and the end coordinates are xc and yc. It is possible then to join these linestogether to form a single poly-line geometry, representing the final spider diagram foreach location using Vector –> Geometry Tools –> Singleparts to Multiparts.

RING TOPOLOGY

The ring topology consists of a single line, connecting the households one by one forminga closed loop. In terms of graphs, every node in this network has a degree of connectivityequal to 2. Also, if any of the edges of the network is removed, the network is still aconnected graph (Section 7.2.1). Households are, therefore, connected to each other withsingle lines, which can then be joined together with the same procedure used for thespider diagram.

BUS TOPOLOGY

Bus topology creation is more complicated. This topology consists of one (or more)straight connection lines (bus lines) running through the village, to which the householdsare directly connected via a (ideally) perpendicular line. To implement that in QGIS, thebus(es) has to be identified and drawn manually. Then, using a short Python script in theconsole, it is possible to identify, for each household, the closest point on the bus line andconnect, with an additional, perpendicular line, the household to the central bus.

RADIAL TOPOLOGY

The radial topology uses the household closest to the centre of the village (xc ,yc ) as acentral starting point for the microgrid. If there is not any household sufficiently central,then the centre of the village itself is used as origin of the network. This also meansthat some of the villages will have a radial network with a number of nodes equal to thenumber of households, while other villages might have one extra hub node as originwhich does not correspond to an actual household, exactly as in the spider diagram. Fromthis origin, the network expands radially towards the most peripheral households. Onceeach radial branch is created, the following households are either connected to the closestbranch, or a new branch is created connecting them directly to the origin, depending onthe distance. For bigger villages, sub-branches can also be added.

MINIMUM SPANNING TREE

Finally, the MST, consisting of the shortest possible combination of edges that leads to acompletely connected graph, is identified in a relatively easy manner in the QGIS envi-ronment using a plug-in available in the QGIS repository: "ReconstructLine". This usefulplug-in accepts as input a set of points, which have to belong to the same shapefile layer,and gives as output a set of single lines connecting one point to the other in a predefinedline shapefile layer, so that the resulting overall network is the Minimum Spanning Treefor that specific set of points, using as working principle the Prim’s algorithm [22]. Thesingle lines can then be merged into a poly-line representing the whole network using thealready mentioned process.

7

154 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

Figure 7.10: Map view and resulting network topologies for location 23. A: Spider Diagram. B: Ring. C: Bus. D:Radial. E: MST.

After the topologies have been created, each of the chosen locations should havefive different layouts to connect its households with each other, without any householdleft disconnected. An example of the graphical visualization of the resulting networks isshown in Figure 7.10 for location 23.

The geometrical length of each of the created network can be easily derived in theattribute field calculator of the line shapefile in QGIS, using the command $length. Spa-tially correlating each of the network to the original village, it is then possible to collect inone single shapefile, namely the original polygon layer in which all the villages were stored,all the information gathered so far: location number, country, area, perimeter, numberof houses, house density and now length of each of the identified topology. It is thentrivial to calculate for each village the average needed length of the network, in [m/hh],directly in the attribute table. A summary of all these parameters and characteristics, forthe considered sample of villages, is shown extensively in Table 7.2 in Section 7.4.

7.3.3. LAYOUT OPTIMIZATIONThe topologies created with the methodology described in Section 7.3.2 are going to beutilized in the following steps as the starting topologies for a layout optimization, whichis going to be performed using an algorithm written in Python.

THE RATIONALE BEHIND THE OPTIMIZATION

So far, the interconnection costs in the form of network length has been the main param-eter of comparison between the topologies. That is why only topologies that are spanningtrees have been selected as the initial topologies for comparison. That is, there are noredundant edges in the topologies (except for the ring topology, which has one extra edgeclosing the ring loop). This ensured a minimization of the network length, within theconstraints imposed by each given topology definition.

7.3. METHODOLOGY

7

155

Nonetheless, cost saving is not the only important aspect while designing a DC mi-crogrid for interconnecting SHS. Additional network lines (added edges), which lead toincreased costs, can bring benefits for other aspects of the microgrid operation. Threeof those aspects are going to be discussed in the following paragraphs: redundancy,congestion avoidance and losses reduction.

Redundancy Redundancy can be interpreted as safety or reliability of the network incase of damages to single lines. Each spanning tree, on losing even a single edge, becomesseparated into two different sub-networks, isolated from each other. This means thatpotential exchange of energy from one sub-network to the other is not possible anymore.Furthermore, if in one of the sub-networks the energy production and energy reserve isnot enough to satisfy the demand of the same sub-network at a given moment, the risk ofenergy outages and the loss of load probability dramatically increases. With additionaledges, increasing the redundancy of the network, additional alternative paths can befound for at least some sub-sections of the network. The most peripheral nodes of thenetwork would always suffer higher risks of isolation then more central, better connectedhouseholds, but, as more edges are added to the network, the overall risk of isolation ofsub-networks is reduced. An example of that can be seen in Figure 7.11.

Figure 7.11: Example of increased redundancy for power flow from A to C by addition of edge AE, making thetopology resilient to a fault in, say, edge BC.

Congestion avoidance Congestion in one or multiple lines is detrimental for the per-formance of microgrid networks. If the power flow exceeds the technical limits of a cable,the consequence can be its failure and hence the complete disconnection of the line.Moreover, the power losses in a cable are proportional to the square of the current flowingin it, leading to very high ohmic losses. As in the case of redundancy, adding new edgesmay offer a second alternative path for power flow from a given household A to a givenhousehold B, reducing risks of congestion. An example can be seen in Figure 7.12.

Line-loss reduction The concept of average shortest path length defined in Section 7.2.1helps in understanding the potential of line loss reduction through the addition of networkedges. The shortest path length in a network for a given pair of nodes is the length of theshortest route to connect those two nodes to each other following the edges of the graph.The average shortest path length is simply the average of this distance over all the possible

7

156 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

Figure 7.12: Example of congestion avoidance by edge addition. Power flow from B to F can be via C as well as Gdue to addition of edge BG.

pairs of nodes in the network. As the cable resistance is proportional to the length of theline, the line losses due to power exchange between households increases with increasingaverage shortest path length. The addition of new edges can create new shortest pathsbetween pair of households, reducing the average shortest path length of the network andpotentially the losses in power exchanges. An example of that can be seen in Figure 7.13.

Figure 7.13: Example of decreased shortest path length between A and E by addition of edge AE.

OPTIMIZATION

The potential benefits of the addition of redundant edges in these topologies necessitatean optimization that not only aims at minimizing cable lengths and therefore cable costsbut also tries to find a trade-off between costs reduction and the said mentioned benefits.It is difficult to quantify these benefits in order to compare them with cable length costs,especially in this first phase of network planning. Nonetheless, in order to comparethese benefits, it was decided to use the average shortest path length of the network as asecond parameter to optimize, taking it as a general indicator of the level of redundancy,congestion avoidance and losses reduction, as opposed to average network length, whichis taken as an indicator of direct cable costs. The proposed target function to be optimizedis shown in Equation 7.5.

f =√wf · (l )2 + (spl )2 (7.5)

in which l is the average network length in [m], spl is the average shortest path length ofthe network in [m] and wf is a user-defined weight factor. This weight factor is tuned to

7.3. METHODOLOGY

7

157

Import Starting Topologyfrom Shapefile

Is there any non­household

node?

Yes

Identify activenodes

Is there a stored best value

for f?

Calculate madd,Max

No

Calculate target functionfnew

Yes

No

is fnew < fbest

No

fnew = fbest G = Gbest

Yes

Has madd,Max been

reached?

Add new edge

No

Remove alladded edges

Yes

No Has tot_iter been reached?

Yes Does the new edge respect the criteria?

Delete Edge

No

YesIs there

another starting topology to import?

Yes

START

Export optimized layout(Gbest ) into shapefile

No

END

Generate additional outputs

Figure 7.14: Flowchart of the proposed algorithm. A starting network topology is imported in .shp format to beprocessed by the algorithm. Passive nodes are identified, if present. Edge addition iterations are performed untiltot_iter is reached, re-initialising the edge addition process from the starting topology every madd,Max edges.Each new edge is checked under a set of criteria before being added. A target function is calculated at everyiteration and the best overall graph Gbest is stored and saved as optimized output. Multiple starting topologiescan be used to optimize a graph (topology).

7

158 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

prioritize one or the other parameter which is considered more important for each singleapplication and target microgrid site.

In this study, different weight factors were used during the analysis to check thesensitivity of the result to weight factor variations. Before microgrid implementation, adeeper analysis is needed to identify a precise case-specific weight factor to be applied,depending on a combination of technical and socio-economical aspects at the microgridsites. Examples of such aspects can be availability of materials, expected energy demand,planned size of SHS, risk of tampering, energy security requirements, climate conditions,and many more. In this work, weight factors in the range of 0.1 to 2 are suggested asreasonable values for the optimization algorithm. The way Equation 7.5 is written, higherweight factors will prioritize cable costs over the mentioned benefits, so minimizationof the average network length will be more relevant than the minimization of averageshortest path length, hence resulting in fewer edges added. On the other hand, lowerweight factors will prioritize redundancy, congestion avoidance and losses reduction,minimizing the average shortest path length as much as possible, hence tending to addmore edges to the starting topology.

The topologies obtained in Section 7.3.2 are going to be used as starting networks forthe addition of new edges, hence forming so called "meshed" microgrids, which can besimply defined as microgrids in which one or more loops can be identified.

ALGORITHM

To perform this optimization, an algorithm in Python language was written, followingthe flowchart shown in Figure 7.14. The algorithm takes as inputs one or more of thestarting topologies, and it processes it by adding new edges for a user-defined number ofiterations, trying to minimize the target function which takes into account both averagenetwork length and the average shortest path length. As some starting topologies mayhave non-household nodes, active (household) nodes are identified first. The maximumnumber of edges mMax for a graph G was given by Equation 7.1. If mstart is the number ofedges in the starting topology, then the maximum number of edges madd,Max that mustbe added to reach a complete graph is given by Equation 7.6, where n is the number ofactive nodes (households).

madd,Max =n · (n −1)

2−mstart (7.6)

As addition of edges till madd,Max can be computationally intensive, a user-defined inputtot_iter is provided such that the iteration stops when this value is reached (eachiteration is an edge added). Overall, the algorithm needs as input the starting topologyshapefile(s), tot_iter for each starting topology, and wf for the target function. Thealgorithm outputs the new shapefile of the optimised graph, a spreadsheet of the l andspl values for all graphs evaluated during the iterations, a scatter graph with those values,and a graphical representation of the network layout. An additionally useful output is aset of .mat files containing information on the nodes and edges of the optimised network.

7.3. METHODOLOGY

7

159

Tab

le7.

2:A

ttri

bu

teta

ble

oft

he

loca

tio

ns

shap

efile

,co

nta

inin

gth

ega

ther

ed,m

easu

red

and

calc

ula

ted

dat

ao

nth

e42

cho

sen

loca

tio

ns.

Loca

tio

nC

ou

ntr

yA

rea

[m2

]Pe

rim

eter

[m]

Ho

use

s[h

h]

Den

sity

[hh

/km

2]

MST

[m]

Spid

er[m

]R

ing

[m]

Rad

ial[

m]

Bu

s[m

]M

ST[m

/hh

]Sp

ider

[m/h

h]

Rin

g[m

/hh

]R

adia

l[m

/hh

]B

us

[m/h

h]

1C

olo

mb

ia11

043.

067

8.0

1412

67.8

362.

587

8.7

564.

436

8.0

418.

625

.962

.840

.326

.329

.92

Icel

and

1182

3.1

618.

06

507.

530

4.0

410.

546

8.0

309.

834

8.3

50.7

68.4

78.0

51.6

58.1

3Ic

elan

d85

486.

216

79.5

893

.682

0.7

1720

.314

13.4

892.

286

3.9

102.

621

5.0

176.

711

1.5

108.

04

Peru

1591

4.7

747.

716

1005

.444

4.6

1176

.469

2.1

456.

554

6.3

27.8

73.5

43.3

28.5

34.1

5Pe

ru79

484.

612

70.4

4050

3.2

1305

.844

03.3

1594

.014

56.1

1962

.132

.611

0.1

39.9

36.4

49.1

6C

olo

mb

ia64

06.2

598.

810

1561

.028

3.9

549.

550

6.0

283.

930

0.5

28.4

54.9

50.6

28.4

30.1

7C

olo

mb

ia11

407.

552

1.0

870

1.3

300.

352

0.4

437.

932

4.8

319.

437

.565

.054

.740

.639

.98

Co

lom

bia

2851

.022

5.7

517

53.8

113.

914

1.0

164.

110

6.9

118.

322

.828

.232

.821

.423

.79

Co

lom

bia

3019

2.6

1059

.512

397.

464

4.7

1328

.093

4.6

717.

476

1.7

53.7

110.

777

.959

.863

.511

Bra

zil

1144

0.3

645.

89

786.

730

9.8

723.

152

4.8

326.

741

6.7

34.4

80.3

58.3

36.3

46.3

10B

razi

l30

608.

213

61.9

1858

8.1

727.

428

06.2

1285

.672

8.8

901.

540

.415

5.9

71.4

40.5

50.1

12C

olo

mb

ia78

70.6

496.

413

1651

.728

1.5

546.

741

4.3

296.

439

8.7

21.7

42.1

31.9

22.8

30.7

13Ta

nza

nia

2471

4.9

1099

.529

1173

.486

9.4

2362

.811

54.5

943.

111

16.5

30.0

81.5

39.8

32.5

38.5

14M

adag

asca

r13

455.

552

0.3

1914

12.1

424.

799

5.2

525.

946

1.5

707.

922

.452

.427

.724

.337

.315

Tan

zan

ia34

537.

613

91.8

1646

3.3

832.

916

11.2

1250

.686

1.2

923.

152

.110

0.7

78.2

53.8

57.7

16Ta

nza

nia

1813

0.1

947.

410

551.

648

9.2

1242

.580

5.1

522.

454

7.5

48.9

124.

280

.552

.254

.717

Tan

zan

ia17

974.

160

4.9

844

5.1

365.

960

1.6

523.

636

5.9

423.

745

.775

.265

.545

.753

.018

Tan

zan

ia52

032.

914

75.2

1528

8.3

863.

825

01.0

1408

.492

1.4

969.

457

.616

6.7

93.9

61.4

64.6

19Ta

nza

nia

7555

.555

6.2

1418

53.0

363.

789

6.5

515.

237

9.9

421.

526

.064

.036

.827

.130

.120

Tan

zan

ia15

026.

563

6.9

1173

2.0

357.

677

1.1

566.

937

0.9

521.

132

.570

.151

.533

.747

.421

Tan

zan

ia86

43.9

466.

58

925.

527

0.2

453.

738

7.3

270.

228

0.7

33.8

56.7

48.4

33.8

35.1

22Ta

nza

nia

4153

9.1

1051

.617

409.

384

2.9

1650

.810

91.1

890.

010

39.0

49.6

97.1

64.2

52.4

61.1

23Ta

nza

nia

6608

.030

1.0

913

62.0

229.

539

3.7

274.

728

8.3

312.

325

.543

.730

.532

.034

.724

Tan

zan

ia14

606.

764

5.3

1495

8.5

426.

810

75.0

652.

645

5.5

487.

430

.576

.846

.632

.534

.825

Nep

al45

111.

116

80.3

3884

2.4

1145

.061

59.5

1688

.511

60.8

1415

.830

.116

2.1

44.4

30.5

37.3

26Ta

nza

nia

2522

9.2

713.

221

832.

454

4.1

1553

.468

7.1

572.

858

6.6

25.9

74.0

32.7

27.3

27.9

27C

amb

od

ia14

5324

.523

85.7

3624

7.7

1979

.476

83.6

2458

.320

83.2

2250

.455

.021

3.4

68.3

57.9

62.5

28Ta

nza

nia

4646

1.8

1234

.620

430.

593

6.2

2520

.012

67.4

959.

311

67.3

46.8

126.

063

.448

.058

.429

Cam

bo

dia

1686

0.5

784.

023

1364

.150

3.8

1897

.369

3.8

510.

579

0.2

21.9

82.5

30.2

22.2

34.4

30In

dia

5705

8.7

1442

.229

508.

295

7.1

3961

.413

39.0

999.

913

52.6

33.0

136.

646

.234

.546

.631

Ind

ia11

606.

258

9.9

2420

67.9

454.

210

74.7

606.

347

1.5

637.

818

.944

.825

.319

.626

.632

Ind

ia11

628.

862

9.4

2723

21.8

461.

017

49.8

561.

946

9.5

587.

017

.164

.820

.817

.421

.733

Ind

ia72

01.9

575.

426

3610

.240

0.7

1050

.354

7.4

411.

048

7.1

15.4

40.4

21.1

15.8

18.7

34In

dia

6877

.849

9.9

1217

44.7

287.

556

3.6

392.

628

7.5

328.

824

.047

.032

.724

.027

.435

Ind

ia51

30.6

378.

720

3898

.228

4.2

653.

638

4.5

293.

933

4.3

14.2

32.7

19.2

14.7

16.7

36B

angl

ades

h31

89.8

310.

416

5016

.020

4.5

437.

723

0.2

212.

623

4.0

12.8

27.4

14.4

13.3

14.6

37B

angl

ades

h68

08.9

370.

420

2937

.332

6.4

633.

841

8.7

335.

544

0.6

16.3

31.7

20.9

16.8

22.0

38B

angl

ades

h33

12.7

385.

910

3018

.717

3.6

367.

625

1.4

173.

620

6.3

17.4

36.8

25.1

17.4

20.6

39B

angl

ades

h60

76.7

417.

524

3949

.531

0.1

835.

342

5.2

329.

337

2.6

12.9

34.8

17.7

13.7

15.5

40In

dia

4266

8.7

1157

.729

679.

799

0.0

3553

.512

90.5

1034

.315

95.5

34.1

122.

544

.535

.755

.041

Cam

ero

on

3126

3.5

1026

.131

991.

684

9.1

3381

.910

16.3

878.

812

25.0

27.4

109.

132

.828

.339

.542

Gab

on

4527

.329

0.1

919

88.0

198.

628

9.6

237.

619

8.9

207.

322

.132

.226

.422

.123

.0

7

160 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

7.4. RESULTS AND DISCUSSION

7.4.1. TOPOLOGY CREATION USING QGISBased on the steps outlined in Section 7.3.2, the 5 different topologies were created foreach of the 42 locations. A summary of all the parameters for each of the 5 topologies forevery location is tabulated in Table 7.2. This is the basis for the comparison of differenttopologies in terms of costs and for the identification of trends useful for evaluating theoptimal design of microgrid topologies.

The household density in the selected locations show a strong correlation with the av-erage network length needed per household to completely connect them into a microgrid.Understandably, the more clustered the households are, the smaller the distance betweenthem. This trend is confirmed by the measured length of all the 5 topologies as shown inFigure 7.15, where the single results for each village and each topology are shown togetherwith topology-specific trend lines. The average lengths range from a minimum value ofup to 14 m/hh for MST topology in really clustered villages, up to a maximum of around215 m/hh for a spider topology in really sparse villages. It is clear, then, how the feasibilityof microgrid is highly affected by the household density and the household distributionof each village location.

0 1000 2000 3000 4000 50000

50

100

150

200

250

300

Household density [hh/km2]

Ave

rage

net

wo

rkle

ngt

h(l

)[m

/hh

] MSTSpiderRingRadialBus

Figure 7.15: Scatter plot showing the variation in average network length l for the different topologies over thehouse densities along with corresponding trendlines.

In terms of general trend, the topologies described can be ranked in the followingorder, from the best to the worst in terms of average network length:MST < Radial Topology < Bus Topology < Ring Topology < Spider Diagram.

However, it must be noted that this ranking and the ratios between different topologiesis not the same for each of the identified villages. In fact, it is dependent on variousfactors, such as the number of houses, the shape of the village and, most importantly,the geometrical distribution of the households inside the village, which can make one

7.4. RESULTS AND DISCUSSION

7

161

topology more convenient for specific cases. Some interesting examples of specificvillages are the following, which are also shown in Figure 7.16.

• Location 7, which has a clear, long shape, is symmetrically distributed on the sidesof a straight road. It does not have any household near the center of the village, andis more suitable for a bus topology (108 m) than for a radial topology (111 m), asseen in Figure 7.16(a).

• Location 23 has a very peculiar circular layout for household distribution. In thissmall village, a ring topology (30.5 m) is slightly more convenient than a radialtopology with the origin in the center of the village (32 m), as seen in Figure 7.16(b).

• Location 8 is the smallest village of the sample. Its shape and the scattered distri-bution of the houses make in this case a Radial topology (21 m) with the origin (afictitious node) in the centre of the village even more convenient than an MST (23m), as seen in Figure 7.16(c).

These are just some of the cases that could be analyzed singularly, but it already givesa hint of how the optimal layout topology is highly case-specific. Nonetheless, it is hopedthat the methodology proposed in this chapter can help in identifying the optimal layoutin such cases.

(a) red = Bus, blue = Ra-dial

(b) red = Ring, blue = Radial (c) red = Radial, blue = MST

Figure 7.16: Specific cases of topology comparison: (a) Location 7 (b) Location 23 (c) Location 8

In general, however, it can be said that the greater the number of houses, the worsethe performance of spider topology. This can be easily explained geometrically as eachhousehold gets connected directly to the centre of the village, creating a number of"branches" which are inefficiently close to each other. Thus, in 40 out of 42 cases in thelocation set, the spider diagram is the least convenient topology. Hence, this particularlayout should be avoided for decentralized microgrid planning. The only exceptionswhere a spider topology might still make sense would be a fully centralized generation orfor centralized public loads in decentralized microgrids.

7

162 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

7.4.2. LAYOUT OPTIMIZATIONThe algorithm described in Section 7.3.3 has been applied to a limited number of villageschosen among the 42. For most of the examined villages, the algorithm was run separatelyfor each different starting topology, obtaining topology-specific optimized meshed gridsas well as overall best solutions. For other villages, especially with higher number ofhouses, the algorithm was run starting only from some of the potential starting topologies,which were considered more suitable, for computational reasons. The smallest villageof the sample set is first considered, which has only 5 households, making it easier tosee how the algorithm works and how the different weight factors used influence theoutcome of the optimization. The values of spl and l for all the potential network layoutsare shown in the scatter graph in Figure 7.17, resulting as an output of the algorithmusing all five different starting topologies. The bigger points with different colours on thetop-left corner of the graph are representing the starting networks.

Figure 7.17: Average network length and average shortest path length of network layouts generated and testedby the algorithm for location 8. Different colours represent different starting topologies.

It can be seen that as more edges are added, thus increasing the average cable length,the average shortest path length decreases, as expected, until it reaches a minimumthreshold. This threshold is reached when the network has become a complete graph,having then the maximum possible average cable length and the minimum averageshortest path length, since every pair of nodes is directly connected via an edge. It isimportant to notice that the complete form of the network changes according to thestarting topology, for example, because of non-household nodes and bus lines, hencechanging the maximum average cable length, but not the minimum average shortest pathlength, which is always reached with spanning trees. This explains why in Figure 7.17 thecomplete graph starting from the spider topology (the farthest on the right pink dot) is

7.4. RESULTS AND DISCUSSION

7

163

not the same as the complete graph starting from the bus topology (the farthest on theright grey dot). It is also interesting to see how the spider diagram, which was the leastconvenient topology in terms of pure cable length (Figure 7.15), is also performing worsethan the other layouts in terms of average shortest path length, reinforcing the statementabout its inadequacy as topology choice for decentralized microgrids.

Figure 7.18: Average network length and average shortest path length of network layouts for location 8, startingfrom MST topology. The green dot indicates the starting network, while the red dot indicates the optimalsolution, found after one edge addition while using a weight factor of 0.3.

The scatter graph resulting from running the algorithm with only the MST as startingtopology is shown in Figure 7.18, which shows how the addition of each edge leads to agradual change in the parameters, from the starting graph to the complete graph. It is alsopossible to graphically see six "stages" of solutions, each representing one consecutiveedge addition. As the number of nodes increases, the computational complexity increasesas well, leading to scatter graphs which are less clear and distinct compared to the simplestcase, but the trends can still be identified. As an example, in Figure 7.19 the scatter graphresulting from running the algorithm for location 12, which has thirteen households, isshown.

INFLUENCE OF WEIGHT FACTOR AND STARTING TOPOLOGY

Going back to location 8, the algorithm was implemented using different weight factorsin the range between 0.1 and 2. As expected, this leads to different optimized layouts,especially in terms of number of added edges. The lower the weight factor, the lowerimportance of network length and the more edges were added to the initial network inorder to reach the optimized topology, as shown in Figure 7.20. It can be seen that forsuch a small village, it can happen that no edges are added at all for some topologiesat high weight factors, because the benefit would be considered too small to justify theaddition of a single edge, hence the optimized layout ends up being the starting topology

7

164 7. OPTIMAL MICROGRID LAYOUT USING GEOGRAPHIC INFORMATION SYSTEM (GIS)

Figure 7.19: Average network length and average shortest path length of network layouts for location 12, whichhas 13 households. Different colours depict different starting topologies.

0.01 0.2 0.5 1 2

0

2

4

6

8

Weight factor [-]

No.

ofe

dge

sad

ded

[-]

BusMSTRingRadialSpider

Figure 7.20: Edges added to reach the optimized layout for village 8 with different weight factors and startingtopologies.

itself. That is for example the case for a weight factor of 2, in which 4 out of 5 startingtopologies do not require any edge addition.

Apart from the single starting topologies, the influence of the weight factor can alsobe seen in the final optimized layouts. In Figure 7.21, the final layout for location 8 isshown for different weight factors. The weight factor 0.01 is willingly exaggerated onthe side of deprioritization of cable length and it was purely chosen to emphasize theedge addition. It can be seen how the layout changes with the weight factor, and alsohow different weight factors can lead to different starting topologies to be optimal. As an

7.4. RESULTS AND DISCUSSION

7

165

(a) wf = 2 (b) wf = 0.3 (c) wf = 0.1 (d) wf = 0.01

Figure 7.21: Global optimized layouts for location 8 with different weight factors.

example, while for a weight factor of 2 the radial topology was a better starting topology,for a weight factor of 0.3 an MST was optimal. It must be also noted that in some casesthe optimal layout can also be reached by starting with different topologies. For example,the optimal layouts for weight factors 0.1 and 0.01 could be reached both starting from aring topology and from an MST topology.

In Figure 7.22, the optimized layouts of village 8 are shown for different startingtopologies. It was chosen to show the topology-specific results for a weight factor of 0.3because all the results had at least one edge added to the starting network. The partialoptimized layouts were then compared to each other to find the global optimal, which inthis particular case resulted to be the one obtained from a MST starting topology.

(a) Original spread (b) MST (c) Radial

(d) Ring (e) Bus (f) Spider

Figure 7.22: Optimised layouts for location 8 with weight factor 0.3 and different starting topologies.

From these observations, it can be said that lower weight factors lead to the an in-creased addition of edges to reach the optimal layout independent of the starting topology.

7

166 REFERENCES

It is also noticed that the edges that are more likely to be added to reach an optimal layoutare usually short edges that close new loops into the network or cut through alreadyexisting loops. As mentioned in Section 7.3.3, the idea of assigning a weight factor is a userchoice and dependent on the specific requirements from the microgrid at the target site.Even though the weight factor has a bearing on the final optimized layout, the generalapplicability of the proposed methodology in this study still stands.

7.5. CONCLUSIONSThis chapter presented a GIS-based methodology to evaluate and compare optimal topolo-gies for decentralized microgrids. The main benefits of the method are the incorporationof ground-level data through GIS and a spatial geometry-based analysis using graphtheory, which enable a passive optimization process at the design stage. The proposedmethodology allows not only for a comparison between various candidate microgridtopologies in terms of network lengths (costs) but also an optimization between cablingcosts and potential operational parameters like line congestion and line losses. This isdone through an algorithm that minimizes a target function comprising average networklength and average shortest path length. In general, denser target locations led to loweraverage network lengths, while graph theory-based layout (MST) outperformed conven-tional topologies like ring, spider, bus in terms of average network length. It was seenthat each location requires case-specific considerations, and case-specific conditionsplay a vital role in the application of this methodology. The user-defined weight factorcan be a useful parameter to tweak while using the proposed methodology.In conclusion,combined use of GIS and graph theory can be powerful in planning of rural electrificationprojects, and classical superimposed microgrid layouts can be outperformed by layoutsobtained from such an integrated methodology.

7.5.1. RECOMMENDATIONS AND FUTURE WORK

The optimization can be made computationally faster through smarter edge addition.The use of weight factors can be better correlated with active power flow analysis thattakes into account the various supply and demand attributes for the microgrid. Theweight factors can also be made a function of the wire gauge chosen.

ACKNOWLEDGMENT

Thanks to Dr. Petra Heijnen for the discussions on Graph Theory and related networkanalysis.

REFERENCES[1] H. Louie, Off-Grid Electrical Systems in Developing Countries. Springer, 2018.

[2] A. Werth, N. Kitamura, I. Matsumoto, and K. Tanaka, “Evaluation of centralized anddistributed microgrid topologies and comparison to open energy systems (oes),” in2015 IEEE 15th International Conference on Environment and Electrical Engineering(EEEIC), pp. 492–497, June 2015.

REFERENCES

7

167

[3] F. Islam, K. Prakash, K. Mamun, A. Lallu, and H. R. Pota, “Aromatic network: a novelstructure for power distribution system,” IEEE Access, vol. 5, pp. 25236–25257, 2017.

[4] D. Burmester, R. Rayudu, W. Seah, and D. Akinyele, “A review of nanogrid topologiesand technologies,” Renewable and Sustainable Energy Reviews, vol. 67, pp. 760–775,2017.

[5] T. Dragicevic, X. Lu, J. C. Vasquez, and J. M. Guerrero, “Dc microgrids — part ii:A review of power architectures, applications, and standardization issues,” IEEEtransactions on power electronics, vol. 31, no. 5, pp. 3528–3549, 2015.

[6] K. Prakash, A. Lallu, F. Islam, and K. Mamun, “Review of power system distributionnetwork architecture,” in 2016 3rd Asia-Pacific World Congress on Computer Scienceand Engineering (APWC on CSE), pp. 124–130, IEEE, 2016.

[7] C. Tíba, A. L. B. Candeias, N. Fraidenraich, E. d. S. Barbosa, P. B. de Carvalho Neto,and J. B. de Melo Filho, “A gis-based decision support tool for renewable energymanagement and planning in semi-arid rural environments of northeast of brazil,”Renewable Energy, vol. 35, no. 12, pp. 2921–2932, 2010.

[8] M. Zeyringer, S. Pachauri, E. Schmid, J. Schmidt, E. Worrell, and U. B. Morawetz,“Energy for Sustainable Development Analyzing grid extension and stand-alonephotovoltaic systems for the cost-effective electri fi cation of Kenya,” Energy forSustainable Development, vol. 25, pp. 75–86, 2015.

[9] F. Kemausuor, E. Adkins, I. Adu-Poku, A. Brew-Hammond, and V. Modi, “Electrifica-tion planning using network planner tool: The case of ghana,” Energy for SustainableDevelopment, vol. 19, pp. 92–101, 2014.

[10] S. Ohiare, “Expanding electricity access to all in nigeria: a spatial planning and costanalysis,” Energy, Sustainability and Society, vol. 5, no. 1, p. 8, 2015.

[11] S. Choudhury, A. Parida, R. M. Pant, and S. Chatterjee, “Gis augmented computa-tional intelligence technique for rural cluster electrification through prioritized siteselection of micro-hydro power generation system,” Renewable Energy, vol. 142,pp. 487–496, 2019.

[12] F. F. Nerini, O. Broad, D. Mentis, M. Welsch, M. Bazilian, and M. Howells, “A costcomparison of technology approaches for improving access to electricity services,”Energy, vol. 95, pp. 255–265, 2016.

[13] F. Dalla Longa, T. Strikkers, T. Kober, and B. van der Zwaan, “Advancing energy accessmodelling with geographic information system data,” Environmental Modeling &Assessment, vol. 23, no. 6, pp. 627–637, 2018.

[14] D. Mentis, M. Andersson, M. Howells, H. Rogner, S. Siyal, O. Broad, A. Korkovelos,and M. Bazilian, “The benefits of geospatial planning in energy access–a case studyon ethiopia,” Applied Geography, vol. 72, pp. 1–13, 2016.

7

168 REFERENCES

[15] D. Mentis, M. Welsch, F. Fuso Nerini, O. Broad, M. Howells, M. Bazilian, and H. Rogner,“A GIS-based approach for electrification planning-A case study on Nigeria,” Energyfor Sustainable Development, vol. 29, pp. 142–150, 2015.

[16] T. Huld, M. Moner-Girona, and A. Kriston, “Geospatial analysis of photovoltaicmini-grid system performance,” Energies, vol. 10, no. 2, pp. 1–21, 2017.

[17] D. Mentis, M. Howells, H. Rogner, A. Korkovelos, C. Arderne, E. Zepeda, S. Siyal, C. Tal-iotis, M. Bazilian, A. de Roo, et al., “Lighting the world: the first application of an opensource, spatial electrification tool (onsset) on sub-saharan africa,” EnvironmentalResearch Letters, vol. 12, no. 8, p. 085003, 2017.

[18] K. R. Varshney, G. H. Chen, B. Abelson, K. Nowocin, V. Sakhrani, L. Xu, and B. L.Spatocco, “Targeting Villages for Rural Development Using Satellite Image Analysis,”Big Data, vol. 3, no. 1, pp. 41–53, 2015.

[19] A. Alhamwi, W. Medjroubi, T. Vogt, and C. Agert, “Development of a gis-based plat-form for the allocation and optimisation of distributed storage in urban energysystems,” Applied Energy, vol. 251, p. 113360, 2019.

[20] M. Brazil and M. Zachariasen, “Optimal interconnection trees in the plane,” Algo-rithms and Combinatorics, vol. 29, 2015.

[21] R. L. Graham and P. Hell, “On the history of the minimum spanning tree problem,”Annals of the History of Computing, vol. 7, no. 1, pp. 43–57, 1985.

[22] R. C. Prim, “Shortest connection networks and some generalizations,” The BellSystem Technical Journal, vol. 36, no. 6, pp. 1389–1401, 1957.

[23] F. Dörfler, J. W. Simpson-Porco, and F. Bullo, “Electrical networks and algebraic graphtheory: Models, properties, and applications,” Proceedings of the IEEE, vol. 106, no. 5,pp. 977–1005, 2018.

[24] S. Nolan, S. Strachan, P. Rakhra, and D. Frame, “Optimized network planning ofmini-grids for the rural electrification of developing countries,” in 2017 IEEE PESPowerAfrica, pp. 489–494, IEEE, 2017.

[25] E. W. Dijkstra, “A note on two problems in connexion with graphs,” NumerischeMathematik, vol. 1, pp. 269–271, 1959.

[26] R. W. Floyd, “Algorithm 97: shortest path,” Communications of the ACM, vol. 5, no. 6,p. 345, 1962.

[27] Quantum GIS, “(software) qgis ver 2.18 las palmas de gran canaria,” 2017.

[28] ArcGIS Map Repository, “Rural electrification densification programme.” Accessed:2018-06-01.

[29] ArcGIS Map Repository, “Hard to reach unions in bangladesh.” Accessed: 2018-06-01.

8CONCLUSION

Everything that has a beginning has an end

The Oracle

8.1. GENERAL CONCLUSIONSThe research work presented in this dissertation was founded on the idea of contributingtowards achieving universal electrification. The existing electrification pathways werecritically analyzed, and the shortcomings of solar home systems, centralized microgrids,and grid extension were discussed. SHS can effectively achieve only lower tiers of electri-fication, viz., up to tier 2 and 3, while centralized microgrids and central grid extensionsuffer from high (techno-economic) thresholds to reach the necessary scale. Researchgaps were identified from a technological point of view, necessitating the work done onadaptable load profile construction as a starting point for off-grid system design. Loadprofiles were constructed comprising dedicated off-grid super-efficient DC appliancesfor each of the five tiers of the multi-tier framework (MTF) for measuring householdelectricity access. Batteries in SHS were another focus of this dissertation, and a batterylifetime estimation methodology was introduced based on dynamic capacity fading. Theproposed lifetime estimation model yielded state of health (SOH) estimations within lessthan 3% of the estimate obtained from an empirically derived model of Lithium-IronPhosphate (LiFePO4) battery over the battery lifetime simulation period. The intricateinterdependencies between SHS parameters led to a detailed system sizing effort takinginto account battery lifetime, excess PV generation, and system power availability fordifferent tiers of electrification. Optimal SHS sizes for each tier of the MTF were obtained,which clearly showed the inadequacy of SHS to meet higher tiers (especially tier 5) ofelectrification. A bottom-up, decentralized, modularly expandable and scalable microgridarchitecture was proposed based on interconnected SHS; a related case-study showedthat more than 40% energy storage sizing gains are achievable for a tier 5 level electricitydemand in the interconnected SHS-based microgrid when compared to standalone SHS.

169

8

170 8. CONCLUSION

Finally, a Geographic Information System (GIS)-based methodology was proposed thatcan identify optimal microgrid layouts while utilizing real-world ground-level data andconcepts from graph theory as applied to network analysis. Case studies of 42 remotelocations for SHS-based microgrids showed that graph theory-based minimum spanningtree (MST) achieved least network lengths when compared to conventional microgridtopologies like bus, spider, and radial topology for most cases. The described integratedmethodology using GIS and graph theory can be applied on a case-specific basis to planrural electrification projects, especially when going from SHS to a bottom-up microgrid.

8.2. CONTRIBUTIONSFollowing are the main contributions of this PhD dissertation.

1. A bottom-up stochastic load profile construction tool that is fully scalable andadaptable to different user groups.

2. A battery lifetime estimation methodology that accounts for dynamic capacityfading of the battery

3. A methodology to optimize the standalone SHS size for various levels of electrifi-cation, taking into account battery lifetime while maintaining adequate levels ofpower supply.

4. An SHS architecture that is modular and scalable for both individual SHS expansionand microgrid expansion.

5. Quantification of the benefits of moving from standalone SHS to sharing energy inan interconnected SHS-based microgrid in terms of battery storage size.

6. A Geographic Information System (GIS)-based, customizable, and region-specificmethodology to identify optimal microgrid topologies based on interconnectedSHS.

8.3. TOPICAL CONCLUSIONSThe key findings that answer the thematic research questions (from Section 1.3.3) arediscussed in this section.

8.3.1. ELECTRIFICATION PATHWAYS

OVERVIEW OF PRESENT ELECTRIFICATION PATHWAYS FOR UNIVERSAL ELECTRIFICATION

Limitations of the present electrification pathwaysThe three different electrification pathways, viz., grid extension, centralized micro-

grids, and standalone solar-based solutions like pico-solar products and SHS, were crit-ically compared on their relative merits and demerits. While standalone solar-basedsolutions like pico-solar and SHS can provide on-demand electrification, they lack thenecessary power levels in their present form to cater to tiers 4 and 5 levels of electricitydemand. Grid extension can easily supply the higher tiers in terms of power level butis far from an on-demand solution to the electrification of remote areas. To a lesser

8.3. TOPICAL CONCLUSIONS

8

171

extent, centralized microgrids also necessitate a minimum population density and de-mand threshold to justify investments. For climbing up the electrification ladder, anSHS-based decentralized microgrid is proposed that could potentially reach higher tiersof electrification without the disadvantages of grid extension and centralized microgrids.

8.3.2. SHSLOAD PROFILES

Constructing load profiles for the various levels of (off-grid) electrificationA stochastic load profile construction methodology was presented in Chapter 3 to

quantify the energy needs for the various tiers of the multi-tier framework in the form ofload profiles. The loads were entirely composed of dedicated, off-grid, and in some casesthe so-called super-efficient, appliances. A method for incorporating the coincidencefactor was introduced, which can help with peak estimation for the load profile. Theload profile estimation methodology is dependent on the sanctity of the appliance levelinputs, which can be enhanced with dedicated field interviews/surveys or even on-linefield data. The load profiles resulting from this work were used as inputs for the rest ofthe dissertation.

BATTERIES IN SHSEstimating battery lifetime for an SHS application

An improved methodology for estimating the battery lifetime was discussed in Chap-ter 4 without needing to model the electrochemical processes within the battery orrequiring dedicated experiments. Four different battery technologies were compared, viz.,Lead-Acid (LA)-gel, LA-flooded, Nickel-Cadmium (NiCd), LiFePO4 for a given SHS tier 3application, with LiFePO4 showing the best results with the highest estimated lifetime of16.7 years. Comparison of this proposed dynamic model with an experimentally derivedempirical model of LiFePO4 battery yielded very close results, with the SOH values overtime being within less than 3% of each other. The methodology outlined in this chaptercan potentially help SHS designers in estimating battery lifetimes at the SHS design stage.The battery lifetime estimation was also used in Chapter 5 for the optimal sizing of SHS.

SYSTEM SIZING

Optimizing SHS size considering battery size, system power availability, excess PV genera-tion, and battery lifetime

Optimal sizing is a crucial exercise in off-grid design as it prevents oversizing orundersizing of SHS, which can lead to high system costs or inadequate power availability,respectively. The study described in Chapter 5 detailed the optimal sizing of SHS for everytier of the multi-tier framework for household electricity access. A genetic algorithm-based multi-objective optimization was performed, which gave insights on the delicateinterdependencies of the various system metrics like loss of load probability, excess energy,and battery lifetime on the SHS sizing. Optimal system sizes for each tier of the MTF werepresented, and the implications of these system sizes were discussed from the perspectiveof state-of-the-art SHS. It was evident that meeting tier 5 level demand was untenablethrough the usage of SHS alone. The performance of the optimal standalone SHS sizeson the system metrics as found in Chapter 5 was used as a baseline for comparing thebenefits of going from standalone SHS to SHS-based microgrid.

8

172 8. CONCLUSION

8.3.3. SHS-BASED DC MICROGRID

BENEFITS OF AN SHS-BASED DC MICROGRID

Quantifying the benefits on the battery storage sizing and power availability when goingfrom standalone SHS to SHS-based microgrids

A modularly expandable and scalable DC microgrid architecture was proposed inChapter 6 as a logical next step for SHS-based electrification. Meeting the energy demandof higher tiers like tier 4 and tier 5 is shown to be possible with this approach of bottom-up,interconnected SHS-based microgrid as opposed to standalone SHS. On modeling theenergy sharing between the SHS, it was shown that battery sizing gains of more than 40%and 30% could be achieved with interconnectivity at tier 5 and tier 4 levels, respectively,as compared to standalone SHS to meet the same loss of load probability threshold.These gains are poised to increase further depending on the demand diversity of the loadprofiles, which tends to increase with increasing microgrid size.

OPTIMAL MICROGRID TOPOLOGIES

Finding the optimal microgrid topology for interconnection of SHS considering the spatialspread of the households

A GIS-based methodology was presented in Chapter 7 to evaluate and compare op-timal topologies for decentralized microgrids. The main benefits of the method are theincorporation of real-world ground-level data through GIS and spatial geometry-basedanalysis using graph theory. The proposed methodology allows not only for a comparisonbetween various candidate microgrid topologies in terms of network lengths (costs) butalso an optimization between cabling costs and operational parameters like line conges-tion and line losses. The selected locations in the case-studies were seen to require adedicated and case-specific consideration. Graph theory-based layout (minimum span-ning tree) outperformed conventional topologies like ring, spider, bus in terms of averagenetwork length. This work underlined the utility of the combined use of GIS and graphtheory to arrive at optimal microgrid layouts for rural electrification planning.

8.4. RECOMMENDATIONS AND FUTURE WORKAt the time of writing this dissertation chapter, the total population without electricityaccess had fallen below 1 billion for the first time since 1990 to 840 million . It is still atall task to electrify the 840 million by 2030. The research conducted in this PhD project,primarily focused on achieving universal electricity access, can be further enhancedthrough the following recommendations.

8.4.1. SHSLOAD PROFILES

The load profile estimation methodology explained in Chapter 3 can tremendously gainfrom field research on load consumption. While the load profile construction can betailored for different target user groups, a back and forth may be needed with the on-fieldusage patterns. This is even more important when considering seasonal patterns or localsocio-cultural behaviours. Comparison of the stochastic load profiles should be made

IEA, IRENA, UNSD, WB, WHO (2019), Tracking SDG 7: The Energy Progress Report 2019, Washington DC

8.4. RECOMMENDATIONS AND FUTURE WORK

8

173

with field data, especially for higher tiers of electrification, as and when such data isavailable.

The deterministic nature of the final load profiles used for the later system analysisalso affects the battery lifetimes and corresponding system sizes. While this does notchange the applicability of the underlying methodologies covered in this dissertation, it isshould be noted that based on the context and site-specific analysis, the actual sizing ofthe SHS components and microgrid might differ from that arrived at in the chapters.

SHS BATTERIES

The work described in the dissertation concerning SHS battery can further benefit fromfield data on on-line SOH monitoring. This can be used to empirically update the lifetimeestimation methodology proposed in Chapter 4 at the SHS design stage. Additionally,more field data from SHS batteries would also help replicate the exact usage conditionsfor battery capacity fading tests in the lab. Hybrid batteries could also be considered tocater to different C-rate requirements, especially for tiers 4 and 5. Autonomous batteryintegrated DC appliances can also be explored to enable a modular growth up the energyladder, especially up to tier 3 level electrification. Battery management systems (BMS) canbe improved upon to accommodate additional capabilities like enabling power-sharingbetween interconnected SHS.

In the analysis considered in this dissertation, the end of life for batteries was consid-ered at 80% state of health as it is usually specified by battery manufacturers. However, forlow C-rate applications like SHS, this limit could be pushed lower, paving the way for theuse of second-life batteries handed down from other applications like electric vehicles.This would need a dedicated investigation on the battery health of second-life batteriesfor SHS applications.

8.4.2. SHS-BASED MICROGRIDS

BOTTOM-UP DC MICROGRID ARCHITECTURE

The dual voltage microgrid architecture suggested in this dissertation leaves several chan-nels open for further research. Having a 350 V DC bus requires additional work on safetyand protection. There were no standards for remote rural DC microgrids available atthe time of writing this dissertation. Voltage set-points for the decentralized controldescribed in Appendix C can benefit from available benchmarks. Having readily availablestandards for the plug-and-play type of (DC) microgrid deployment can also greatly in-crease the implementation and adoption levels of decentralized (DC) microgrids. Thenon-availability of DC microgrid standards for rural electrification for above 100 V, cou-pled with slow proliferation of high power DC appliances for productive usage will keepsolar-AC systems very much in competition in the foreseeable future. Also, further re-search is needed to compare the topologies obtained through the GIS study in Chapter 7in terms of experimental implementation.

ALGORITHMS FOR ENERGY EXCHANGE

Research into smart algorithms is needed for energy exchanges in such remote rural DCmicrogrids. The algorithms should enable the decentralized control and expansion ofsuch microgrids, while also maintaining the levels of robustness needed in the microgrid

8

174 8. CONCLUSION

for deployment and adoption in low-resource settings. Additionally, the algorithms shouldfacilitate energy exchange while minimizing line congestion in the microgrid. Finally,monetization of energy exchange for such a network should be explored, so that positivecash-flows can arise in such community microgrids, thereby lessening the burden onOPEX expenditures for maintenance and upkeep of the microgrid infrastructure.

8.4.3. MULTIDISCIPLINARY RESEARCHFinally, the technical research conducted in this project and presented in this dissertationis only part of the story when it comes to the larger target of universal electrification.Although not within the scope of this dissertation, there are several other areas of researchthat need to be undertaken to make significant strides towards universal electrification.For instance, innovative business models are needed to make the SHS-based technologicalsolutions accessible and affordable to the most energy-deprived communities. Context-specific user research needs to take place, especially when productive use of energyneeds to be taken into account. Additionally, policies need to be updated and adapted toaccommodate as well as aid the rapid rise of SHS with their batteries and DC appliances.Policies will also be instrumental for supporting the growth of DC microgrids that canutilize existing SHS as well as be ready for eventual grid expansion.

Thus, to fully assess the feasibility of such an SHS-based scalable off-grid electrifi-cation vision towards universal electrification, a broader multi-disciplinary approach isneeded that includes fitting business models, socio-cultural aspects of contexts where themicrogrid is to be deployed, and effective policies that stimulate technology adoption.

Each day we go to our work in the hope of discovering — in the hope that some one, nomatter who, may find a solution of one of the pending great problems, and each succeedingday we return to our task with renewed ardor; and even if we are unsuccessful, our work

has not been in vain, for in these strivings, in these efforts, we have found hours of untoldpleasure, and we have directed our energies to the benefit of mankind.

Nikola Tesla

APPENDIX A: CONSTRUCTING

EQUIVALENT ELECTRICAL CIRCUIT

MODELS OF BATTERY CELLS FOR

SHS APPLICATIONS

A.1. INTRODUCTIONIn this study, two battery technologies commonly used in SHS were explored: the valveregulated lead–acid (VRLA) battery and the LiFePO4 (LFP) battery. Lead–acid is the cheap-est technology in the market in comparison with the other common battery technologies[3]. In particular, the valve regulated type of lead–acid battery is sealed, which reducesthe risk of liquid leaking when compared to flooded lead–acid battery. Moreover, VRLAbattery is entirely maintenance free [4]. In the case of LFP battery, it has surged as apredominant technology due to its appropriate chemical and thermal stability togetherwith its high energy capacity [5]. Additionally, LFP batteries have experienced a reductionin price that is expected to continue [6].

A.1.1. SELECTING THE BATTERY MODEL

Depending on the application, accuracy, complexity, compatibility and universality, thereare several technical methods to approach an expected battery model. Generally, thereare three levels of battery models according to the degree of physical insight, viz., whitebox (e.g., electrochemical/physical model), grey box (e.g., equivalent electrical circuitmodel) and black box (e.g., artificial neural network model) [7].

The equivalent electrical circuit model (EECM) presents a good balance of compu-tational time, precision, and complexity. As the EECM models the battery with onlyelectrical components, they are highly compatible with other electrical models, whichmake them easy to implement with battery management systems and in other electricalengineering applications. Moreover, the construction of EECM does not have a verycomprehensive co-relation with electrochemical processes inside the battery. As a result,there is no limitation on distinct battery types when modelling the battery by EECMs.Hence, the electrical equivalent circuit model (EECM) is selected in this study.

This appendix chapter is based on the following publication: Yu, Y.; Narayan, N.; Vega-Garita, V.; Popovic-Gerber,J.; Qin, Z.; Wagemaker, M.; Bauer, P.; Zeman, M. Constructing Accurate Equivalent Electrical Circuit Models ofLithium Iron Phosphate and Lead–Acid Battery Cells for Solar Home System Applications. Energies 2018, 11(9),2305.

175

176 APPENDIX A

A.1.2. IMPORTANCE OF LOW CURRENT BATTERY MODELS FOR SHSBattery modelling is essential in SHS design, as they help to elucidate the relationshipbetween power required by a predefined load and the operational conditions in whichthe batteries are used. Particularly, in SHS, the C-rates (related to the rate at which abattery is charged or discharged) are usually lower than 0.5 C, with the lowest C-rate being1/120 C to 1/100 C [8, 9].However, the existing battery cell level models do not focus onlow C-rates; instead, they normally study a wide range of C-rates with high samplingresolution. For example, C-rates of 0.1 C, 0.5 C, 1 C, 2 C, 3 C, 4 C, 7 C, and 9 C have beenproposed in [10], while other authors have explored C-rates of 0.04 C, 0.1 C, 1 C, 2 C, 3 C, 5C, and 10 C [11], and 0.1 C, 0.5 C, 1 C, 2 C, 3 C, and 4 C [12]. Therefore, the applicability ofthese models in wide C-rate ranges does not necessarily validate their suitability for SHS.

Consequently, we focus on the low battery C-rates reflecting typical SHS applications.To obtain an accurate battery model which suitable for low C-rate applications, experi-ments with narrower C-rate range but with high sample resolutions are proposed. Theapplicable C-rate range of this model is 0.05 C–0.5 C with the sample resolution of 0.05C–0.1 C.

A.1.3. CONTRIBUTIONSThe work described in this appendix chapter contributes to the current state-of-the-art by

• developing an accurate battery cell level model especial for a low currents, as thecase in SHS;

• proposing a common modelling methodology based on electrical circuit modelapplicable to both VRLA and LFP batteries; and

• improving the current electrical circuit models for VRLA battery including

– a non-linear relation between VOC and SOC;

– a 2nd order RC circuit-based EECM model using a Thevenin approach;

– considering the parasitic branch of the EECM in terms of SOC and C-rate-based Coulombic efficiency.

A.2. BACKGROUND

A.2.1. BATTERY PARAMETERSState of the Charge (SOC): SOC is usually defined as the proportion of the maximumpossible electric charge that is present inside a rechargeable battery [13]. It is calculatedas the ratio between the difference of the battery capacity and the net electric chargedischarged from a battery since the last full state of charge. The SOC can be expressed byEquation A.1.

SOC = Qt −Qe

Qt= Qr

Qt(A.1)

Qe =∫ t

0Imdτ (A.2)

A.2. BACKGROUND 177

Im = Ibatt − Ip (A.3)

where Qt is the temperature dependent actual capacity, Qe is the capacity that be extractedfrom the battery starting from its full state, and Qr is the capacity that remaining in thebattery. Im is the main branch current, which is the result of the load current applied tothe battery minus the current consumed in side-reactions (Ip). Im is positive if the batteryis discharging and Im is negative if the battery is charging.

Electromotive force (emf): The electromotive force of a cell is the algebraic sum of thepotential difference of two half cells, which is the chemical potential of redox reactionshappening on these two electrodes [14].

Open circuit voltage (OCV, or VOC) and cte-OCV : The open circuit voltage (OCV) isthe potential difference across its terminals when there is no current flow in or out of areversible cell [15]. If the battery rests enough time (depending on the battery technology)with no current flow, the OCV can be called Close-to-Equilibrium open circuit voltage(cte-OCV) [16]. This value can be regarded as the estimated emf value.

Current rate (C -rate): The current rate, which is also called C -rate, is defined as theratio of magnitude of charge/discharge current in Amperes to the nominal capacity of thebattery in Ampere-hour. The C-rate represents the speed of charge/discharge, and it canbe calculated using Equation A.4 [17].

C − r ate = Ibatt

Qn(A.4)

where Ibatt is the current applied on the battery in the unit of ampere (A), and Qn is thenominal battery capacity in the unit of ampere hour (Ah).

A.2.2. CONSTRUCTION OF THE DYNAMIC BATTERY MODEL

A good battery model should be capable of predicting both the battery storage capabilityinformation and the voltage response to the load. An equivalent electrical circuit model(EECM) can describe both characteristics. In some cases, modelling the side reactions isrequired in terms of battery losses, which is also possible to be realized with the EECM.

As shown in Figure A.1, there are three main parts in the EECM to model the differentfeatures of the battery. First part is the storage circuit. Three parameters should bemodelled in this sub-circuit: the battery capacity (which is the charge storage ability), theamount of charge is left in the battery (SOC), and the open circuit voltage VOC.

The second part is the voltage response circuit, which focuses on the internal structureof the battery. This makes it possible to imitate the electrical response of the battery underdifferent loads. In this study, a Thevenin model is applied, and it consists of a resistancein series with RC pairs.

The last part is the parasitic branch, which is used to model the side reactions thatinduce loss of charge in the battery. There are many ways to model these battery losses.For example, Figure A.1 shows the parasitic branch in the form of a resistance (Rp). Forboth battery technologies, the basic construction of EECM is applicable, as shown inFigure A.1.

178 APPENDIX A

Figure A.1: Schematic of the overall EECM construction.

A.2.3. STORAGE CIRCUITAs discussed in Section A.2.1, the cte-OCV can be regarded as the emf value. As the emfvalues are SOC dependent, the cte-OCV can be used as a SOC indicator. In this study, theSOC value is selected as the independent variable, while the coulomb counting method isused to determine the SOC. The SOC can be expressed by Equation A.1.

A.2.4. ELECTRICAL RESPONSE CIRCUITThere is a deviation of voltage from the equilibrium state, also called polarization, when-ever the battery is loaded. Conversely, the voltage recovers back towards the equilibriumstate whenever the load is disconnected. Hence, a voltage relaxation phenomenon isobservable in a no-load interval (also called rest interval) when the battery is operating.The battery internal impedance can be potentially extracted from this voltage relaxation.

The non-load rest interval during a discharge stage is shown in Figure 2(b). Thevoltage response can be divided into two parts, viz., an instantaneous voltage jump rightafter the load is paused, followed by a gradual increase of the voltage towards the cte-OCV.Those two voltage responses are correlated with two electrical parts: the ohmic resistanceRo, and the resistance/capacitance pairs (R1C1, R2C2), as introduced in the constructionof EECM model in Figure A.1.

Generally, there is an expected time window for each of the voltage response. AnEECM containing two RC-pairs is assumed. It is suggested that, within a five secondstime-window, the voltage boost (Vo) right after the load disconnection is induced by theohmic resistance Ro [18]. The time window for the rest of the gradual voltage increasecaused by the RC pairs is called the time constant. The time constant reflects the rateof change of voltage; if the voltage changes faster, the time constant is smaller [19]. Theincrease in voltage right after the immediate voltage increase is the “short-term voltageresponse”. This short time interval (τ1) is expected to be around 10 seconds, and thisvoltage increase (V1) is related to the first RC-pair, which is the R1, C1 in Figure 2(b) [20].The following increase in voltage (V2) is then treated as the long-term voltage responseafter a long time interval (τ2), which corresponds to the second RC-pair (R2, C2), as shown

A.2. BACKGROUND 179

(a) The electricalcircuit elements

0 50 100 150 200 250 300 350 400 450 500Time [s]

3.55

3.555

3.56

3.565

3.57

Vol

tage

[V]

0

0.1

0.2

0.3

0.4

0.5

0.6

Cur

rent

[A]

VoltageCurrent

Rest intervalV

initial

OCV

Vo

12

V1

V2

Vinterval

(b) The voltage relaxation curve

Figure A.2: An example of voltage relaxation during rest interval in discharge with corresponding circuitelements.

in Figure 2(b). It is possible to increase the number of RC-pairs and divide the voltagerelaxation into more parts, with which the accuracy of the model is improved, such asthe Rn, Cn in Figure 2(b). Many authors suggest to model the lead–acid battery with firstorder RC circuit. However, for Li-ion batteries, all three orders of RC circuit model arecommonly seen [21, 12, 22, 23, 24]. Eventually, the relaxation voltage will reach cte-OCVif the rest interval is long enough. However, it is necessary to strive an optimum balancebetween accuracy and complexity. Therefore, the study described in this chapter hasconsidered two RC pairs.

Curve fitting method was applied on each of the voltage relaxation results. Therecognition of each part of the voltage and the corresponding time constant was done bynonlinear least squares curve fit in Matlab [20]. The distinct voltage and time constantvalues were then fitted into the RC circuit complete response equations to obtain thevalues of electrical elements in the model. The equations are listed below as Equations A.5to A.9.

Vinterval =V1(1−e−tτ1 )+V2(1−e

−tτ2 )+Vo (A.5)

VOC =Vinitial +Vo +V1 |t→∞ +V2 |t→∞ (A.6)

Ro = Vo

I(A.7)

Rk =Vk

I(A.8)

Ck =τk

Rk(A.9)

180 APPENDIX A

where k=1, 2, 3, ..., n.

A.2.5. PARASITIC REACTION CIRCUITThe electrical components interpreted in the previous subsections describe the behaviourof the battery based on its main reactions. During operation, not all the battery energy isinvolved in its main (charging/discharging) reactions; there are unwanted side reactionsresponsible for loss of useful energy. Therefore, additional elements are needed to modela parasitic branch considering the side reactions.

Especially when charging a lead–acid battery, significant losses can occur, which has ahuge impact on the battery dynamic behaviour. This is particularly true when the batteryis at a high SOC (>95%). Therefore, the charge loss of VRLA during the charging stage isconsidered as the most dominant loss mechanism. This loss is mainly due to gassing andoxygen recombination, but the transient gassing phenomenon can be quite complex andunpredictable especially for VRLA batteries [25].

There are different methods to model the parasitic branch. Some models use aself-discharge conductance to represent the side reactions, where the conductance isthe function of battery voltage, or side reaction related voltage and the temperature[21, 22]. Some other models represent the side reaction by means of gassing current, asthe function of water decomposition voltage and the temperature, instead of modellingthe side branch circuit [26].

In this study, an equation is derived from the one proposed in [27]. This method isbased on the principle that the losses incurred in the parasitic branch can be viewed asthe Coulombic inefficiency of the battery. Therefore, this method models the parasiticbranch in the form of the Coulombic efficiency as a function of SOC and the batterycurrent, as shown in Equation A.10.

ηc = 1−exp

(b

a × ItestIref

+ c× (SOC −1)

)(A.10)

where Iref is the reference current, and Itest is the test current. a, b, and c are the exper-imentally derived empirical constants. It must be noted that, as the current terms inEquation A.10 occur in the form of a ratio, the Coulombic efficiency can be considered asa function of C-rate as well.

A.3. METHODS AND EXPERIMENT

A.3.1. CHOICE OF OPERATIONAL VARIABLESFor both battery technologies, each of the internal impedances is known to be, to someextent, influenced by the SOC, the temperature, and the current [28]. To obtain a properdynamic battery model, a battery should ideally be tested with all three variables. However,the influence of operating temperature was not investigated in this study. All experimentswere performed in a laboratory with the temperature controlled in the range of 20–22C. Therefore, all experiments can be considered as under constant room temperatureconditions. SOC and current (in the form of C-rate) were the main operational variablesconsidered in this study.

A.3. METHODS AND EXPERIMENT 181

A.3.2. OVERALL METHODOLOGYThe EECM model is assumed as shown in Figure A.1, with two RC pairs. The choice oftwo RC pairs is justified in Section A.4. For the task of fully constructing the EECM modelfor both VRLA and LFP batteries, the following methodology was followed involvingexperiments, data analysis, and model verification.

1. Experimentation. Series of experiments were performed to extract the parametersof each electrical element pertaining to every sub-circuit.

(a) Storage circuit. The cte-OCV measurement method was performed.

(b) Voltage response circuit. Voltage relaxation (also called step response) methodwas used.

(c) Parasitic branch. The differential recharge efficiency measurement methodwas used.

These measurements were applied on each featured SOC and with selected currentrate, as further described in Table A.1.

2. Parameter extraction. The parameters of each component in the EEMC wereextracted and analyzed based on the experiments. The values of those parameterswere then summarized into equations, and those equations can represent each ofthe electrical element in the electrical circuit.

3. EECM construction and model verification. All parts of the electrical elementswere put together to construct the full EECM. Finally, the constructed EECM of eachbattery was simulated and compared with the experimental results.

A.3.3. EQUIPMENT AND MATERIALSIn this study, the battery tester used is MACCOR® Series 4000 Automated Test System [29].The batteries selected for tests are Cyclon® AGM D single cell and A123systems® ANR26650M1B,and both of the battery cells have 2.5 Ah capacity [30, 31].

Table A.1: Specification and summarized charge/discharge method for both battery technologies.

Battery Brand A123systems® APR26650M1B Cyclon® AGM D Single Cell

Battery capacity 2.5 Ah 2.5 AhNominal voltage 3.3 V 2 VEnd of charge CC-CV: 3.6V until I ≤ 0.01C CC-CV: 2.5V until I ≤ 0.002C

End of discharge CC-CV: 2.5V until I ≤ 0.01C

Depending on C-rate and EODV:C-rate 0.05 0.1 0.2 0.4 1 2 >5EODV 1.75 1.7 1.67 1.65 1.6 1.55 1.5

The basic charge and discharge methods applied in this study are constant current(CC) method and constant current–constant voltage (CC-CV) method. The CC method in-volves charging or discharging the battery with a constant current until the battery voltagereaches a specific value. The specified voltage to stop the discharging process is called theend of discharge voltage (EODV). The CC-CV method involves first charging/dischargingthe battery with constant current until the battery voltage reaches the pre-set value, and

182 APPENDIX A

then switch to constant voltage stage, which involves charging/discharging the batterywith a constant voltage.

The in-detailed charge and discharge method of each battery technology is summa-rized in Table A.1. This follows the recommendations and best practices outlined in thedata sheets and previous articles [30, 31, 32, 20].

A.3.4. STORAGE CIRCUIT

The purpose of this set of experiments is to obtain the relation between the cte-OCV andthe SOC. Therefore, the cte-OCV should be measured at different SOC values, and thebattery is fully discharged before the measurement. During the experiments, the batterywas monotonically charged to full and then discharged to empty, with both proceduresbeing executed in steps of 5% SOC. Coulomb-counting method was used to approachthe designated SOC demarcation. After each step, the battery was rested for a constanttime to reach the close-to-equilibrium state, at which point the voltage was measuredand noted as the cte-OCV value. The C-rate was 0.2, while the rest time was 3 h for VRLAbattery and 4 h for LFP battery. Three battery cells were tested to get a result with betteraccuracy.

The experimental procedure is illustrated in the flowchart shown in Figure A.3.

Figure A.3: Test procedure of cte-OCV-SOC measurement and the Internal impedance circuit measurement.

A.3.5. VOLTAGE RESPONSE CIRCUIT

The electrical elements in the model were tested while the battery underwent eightdifferent C-rates at different SOC levels. The test procedure esd similar to the cte-OCVtest, as shown in Figure A.3, with the only difference being the amount of energy beingcharged/discharged in each step, and the resting time interval. In this test, the step size ofcharge/discharge was 0.15 Ah, and the relaxation time between each step was 3 min forboth battery technologies. The resting time between charge/discharge was 1 h for bothbattery technologies.

The C-rates tested were: 0.05 C, 0.1 C, 0.15 C, 0.2 C, 0.25 C, 0.3 C, 0.4 C, and 0.5 C.

A.3.6. PARASITIC BRANCH

A simplified measurement was applied to estimate the Coulombic efficiency in this study.The Coulombic efficiency was approximately measured and calculated from the integralrecharge efficiency. The test procedure is outlined as follows.

A.3. METHODS AND EXPERIMENT 183

1. Fully charge and discharge the battery to measure the initial battery capacity andthe overall Coulombic efficiency.

2. Discharge the battery to a specific SOC value (x%) and note down the dischargedcapacity Qd.

3. Fully recharge the battery and record the recharged capacity Qc.

4. Calculate the recharge efficiency Qd/Qc assuming it is the averaged integral rechargeefficiency at the mean SOC between x% and 100%.

5. Go back to Step 2 and repeat the steps until enough data points are obtained.

6. Test the whole procedure under different C-rates.

One thing should be noted: after each charge/discharge step, the battery needed torest until reaching its equilibrium state, for 1 h in this case.

Then, the averaged integral recharge efficiency was fitted to Equation A.10 to get theroughly estimated Coulombic efficiency as a function of SOC and C-rates.

The experimental settings are summarised in Table A.2.

Table A.2: Experimental setting for Coulombic efficiency measurement for VRLA battery.

High SOC Low SOC

SOC value 90% to 99%; 1% steps 10% to 90%; 10% stepsC-rate 0.1 C, 0.2 C, 0.3 C, 0.4 C 0.1 C, 0.2 C, 0.3 C, 0.4 C

A.3.7. EECM CONSTRUCTIONAfter achieving all the required values of the electrical elements, the battery model can beconstructed according to the EECM as shown in Figure A.4.

It must be noted that the parasitic resistance Rp shown in Figure A.4 is not an actualcircuit element in the EECM but only a notional lossy element that signifies side-reaction.In the proposed model in this study, the VRLA losses are actually quantified through theCoulombic efficiency, as shown in Equation A.10.

The SOC of the battery can be represented by Equation A.1, while the main branchcurrent Im is calculated through Equation A.11. The Coulombic efficiency ηc is obtainedfrom Equation A.10.

Im = ηc × Ibatt (A.11)

The dynamic battery voltage of each part of the circuit can be calculated by using theachieved electrical elements through simple circuit analysis, as shown in Section A.2.4.The battery voltage is shown in Equation A.12.

Vbatt =VOC ±Vo ±Vs ±Vl (A.12)

184 APPENDIX A

Figure A.4: The construction of the EECM. The subscripts s and l stand for short and long time durationrespectively with respect to voltage relaxation (τ1, τ2 in Figure 2(b)).

where Vbatt is the battery voltage and the ± sign in Equation A.12 is dependent on thecurrent direction, i.e., + implies charging and − implies discharging the battery.

A.4. RESULTS AND DISCUSSION

A.4.1. EXPERIMENTAL RESULTS

CTE-OCV AS A FUNCTION OF SOCThe cte-OCV and SOC are correlated; as the SOC increases, the cte-OCV also rises. Al-though many papers and manufacturers’ datasheets suggest that the cte-OCV of VRLAversus SOC shows a linear relation [33, 34], our study found that a rational curve fit isbetter than a linear fit. This new fit is especially relevant when the battery is at low SOC,as plotted in Figure A.5. Hence, the empirical-based rational Equation A.13 was proposedand employed instead of a linear relationship.

VOC = 0.2428×SOC 2 +1.935×SOC +0.09876

SOC +0.05336(A.13)

In the case of LFP battery (Figure A.6), there is a sudden increase in VOC when the SOCis close to 100%; similarly, the VOC drops when the SOC is almost 0%. The cte-OCV andSOC behaviour can be perfectly fitted into a double exponential curve [35]. This has beenverified by this study, as shown in Figure A.6, where the hysteresis phenomenon betweenthe VOC values when charging or discharging is depicted. This hysteresis phenomenonwas modelled in the electrical response circuit instead of VOC-SOC equation for bothbatteries.

The fitted cte-OCV-SOC result of LFP battery is written in Equation A.14.

VOC = 3.307×e−0.004117

SOC+0.01772 +e138.7(SOC−1.013) +0.05098×SOC (A.14)

INTERNAL IMPEDANCE

To choose an optimised EECM structure, the fitting comparison from first to third ordersof RC circuits are plotted in Figure 7(a) and 7(b). These two figures show one sample

A.4. RESULTS AND DISCUSSION 185

0 0.2 0.4 0.6 0.8 1SOC

1.8

1.85

1.9

1.95

2

2.05

2.1

2.15

2.2

VO

C [V

]

VOC-c

Cell 1

VOC-c

Cell 2

VOC-c

Cell 3

VOC-d

Cell 1

VOC-d

Cell 2

VOC-d

Cell 3

(a)

0 0.2 0.4 0.6 0.8 1SOC

1.8

1.85

1.9

1.95

2

2.05

2.1

2.15

2.2

VO

C [V

]

Mesured valueRational fitLinear fit

(b)

Figure A.5: Measured open circuit voltage versus SOC of VRLA battery: (a) VOC measured during charge anddischarge separately; and (b) the measured data points and the fitted curves.

of voltage increase during the relaxation time and the fitting result with three differentorders of RC circuits for both battery technologies.

From these curves, it can be concluded that the second order RC circuits have suffi-cient accuracy for both battery technologies. In addition, modelling the battery with ahigher order of RC circuit requires more parameters. Thus, the second order RC circuitswere selected for modelling both battery technologies owing to their lower complexitybut sufficient accuracy.

The measured and fitted results of electrical elements in EECMs of LFP and VRLAbatteries are shown in Figures A.8–A.11, where Ro is the ohmic resistance; Rs, Cs are thevalues of RC pair representing the voltage relaxation in the short time-window (τ1 inFigure 2(b)); Rl, Cl are the values of RC pair used to represent the voltage relaxation in thelong time-window (τ2 in Figure 2(b)).

In Figures A.8 and A.9, some features of VRLA batteries can be inferred with respect tothe internal impedance circuit elements, as listed below.

1. All the electrical elements show a strong dependency on SOC.

2. All the elements also show an inverse trend versus SOC in charge stage and dischargestage.

3. In terms of current dependency, elements during charge and discharge stages havediverse behaviours. While charging, only Rl and Cl values show a clear operationalcurrent dependence. While discharging, all the elements except Ro are evidentlyinfluenced by the operational current.

Therefore, the Rl and Cl in discharge stage and all the elements in charge stage weremodelled as the function of operational current. The detailed equations are listed asbelow:

186 APPENDIX A

0 0.2 0.4 0.6 0.8 1SOC

2.6

2.8

3

3.2

3.4

3.6

VO

C [V

]

VOC-c

Cell 1

VOC-c

Cell 2

VOC-c

Cell 3

VOC-d

Cell 1

VOC-d

Cell 2

VOC-d

Cell 3

(a)

0 0.2 0.4 0.6 0.8 1SOC

2.6

2.8

3

3.2

3.4

3.6

VO

C [V

]

Measured valueFit result

(b)

Figure A.6: Measured open circuit voltage versus SOC of LFP battery: (a) the hysteresis of VOC when charge anddischarge the battery separately; (b) the measured data points and the fitted curve.

CHARGE

Ro_VAc = 0.03782 ·e1.301·SOC +4.226e −6 ·e12.64·SOC (A.15)

Rs_VAc = 4.246e −13 ·e29.05·SOC +0.03787 ·e1.467·SOC (A.16)

Cs_VAc =−2666 ·SOC 4 +4113 ·SOC 3 −1558 ·SOC 2 −112.6 ·SOC +229.3 (A.17)

Rl_VAc =(4.952 ·SOC 4 −7.819 ·SOC 3 +4.168 ·SOC 2 −0.8406 ·SOC +0.1783)

× (9.8198e −4 ·e−48.992·(Cr−0.2) +0.67172 ·e−5.6015·(Cr−0.2) +0.3273

)× (−2.191 ·SOC 3 +2.62 ·SOC 2 −1.006 ·SOC +1.12

)(A.18)

Cl_VAc =[−8768 · (SOC −1)3 −1.961e4 · (SOC −1)2 −1.512e4 · (SOC −1)

]× [

11.66 · (Cr −0.2)3 −4.366 · (Cr −0.2)2 +2.43 · (Cr −0.2)+1]

(A.19)

DISCHARGE

Ro_VAd =0.01226 ·SOC 3 +0.007354 ·SOC 2 +0.02339 ·SOC +0.003575

SOC +0.03138× [−78.342 · (Cr −0.2)4 +25 · (Cr −0.2)3 +4.6032 · (Cr −0.2)2 −2.1016 · (Cr −0.2)+1

]× (

0.4411 ·e−16.19·SOC +0.9853 ·e0.01733·SOC )(A.20)

A.4. RESULTS AND DISCUSSION 187

0 50 100 150 200Time [s]

0

0.01

0.02

0.03

0.04

0.05

0.06

Vol

tage

[V]

Measured voltage response points1st order RC circuit2nd order RC circuit3rd order RC circuit

Adj R-sq: 0.9834

Adj R-sq: 0.9987

Adj R-sq: 0.9156

(a)

0 50 100 150 200Time [s]

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

Vol

tage

[V]

Measured voltage response points1st order RC circuit2nd order RC circuit3rd order RC circuit

Adj R-sq: 0.9985

Adj R-sq: 0.9999

Adj R-sq: 0.9832

(b)

Figure A.7: Comparison of accuracy with modelling battery into different orders of RC circuits: (a) VRLA battery;(b) LFP battery.

Rs_VAd =0.01341 ·SOC 3 +5.996e −4 ·SOC 2 +0.02 ·SOC +6.888e −4

SOC +0.01043

×[(6.6819e2 ·e−0.0011936·(Cr−0.2) +0.15861 ·e−14.305·(Cr−0.2) −667.3473

)×1.003 ·e0.01652·SOC +0.3767 ·e−20.89·SOC ] (A.21)

Cs_VAd =(−1116 ·SOC 4 +2320 ·SOC 3 −1640 ·SOC 2 +458.5 ·SOC +49.67)

× [(−3.3295 · (Cr −0.2)2 +2.7116 · (Cr −0.2)+1)× (

0.9504 ·e0.0464·SOC +0.3536 ·e−609.2·SOC )](A.22)

Rl_VAd =0.1373 ·SOC 4 −0.1344 ·SOC 3 +0.04953 ·SOC 2 +0.01366 ·SOC +0.003415

SOC +0.01673

×[(0.13707 ·e−16.624·(Cr−0.2) −1.021 ·e0.91595·(Cr−0.2) +1.8839

)×0.9371 ·e0.05663·SOC +0.421 ·e−13.09·SOC ] (A.23)

Cl_VAd =(−6879 ·SOC 4 +1.73e4 ·SOC 3 −1.877e4 ·SOC 2 +8425 ·SOC +492.4)

× [3.8298 · (Cr −0.2)+1]× (0.5753 ·e−674·SOC +0.8458 ·e0.2202·SOC )

(A.24)

From the results in Figure A.10 and A.11, some conclusions for LFP batteries can bedrawn:

188 APPENDIX A

0 0.2 0.4 0.6 0.8 1SOC

0

0.5

1

1.5

Ro v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC0

0.5

1

1.5

2

Rs v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

100

200

300

Cs v

s fit

[Far

ad]

0 0.2 0.4 0.6 0.8 1SOC

0

0.5

1

1.5

Rl v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

2000

4000

6000

8000

Cl v

s fit

[Far

ad]

C 0.05 0.05 fit C 0.1 0.1 fit C 0.15 0.15 fit C 0.25 0.25 fit C 0.3 0.3 fit C 0.4 0.4 fit C 0.5 0.5 fit

b)

d)

e)

c)

a)

Figure A.8: The values of RC elements and their fitted results: VRLA batteries during charge.

1. Except for Ro, all other electrical elements show a strong dependency on SOC.

2. All electrical elements exhibit an inverse behaviour versus SOC in charge and dis-charge.

3. The operational current influences Rs, Rl and Cl in both charge and discharge pro-cesses.

Therefore, all the elements could be expressed as the function of SOC for betteraccuracy, even though Ro does not vary too much with different SOC values. Elements Rs,Rl and Cl were modelled as the function of both SOC and C-rate.

In these two figures, it can be observed that the fitting of all the electrical elementsfor battery model construction has a high accuracy. The equations of all these electricalelements for LFP battery are listed below.

CHARGE

Ro_LPc = 0.0334·SOC 4−0.06141·SOC 3+0.03985·SOC 2−0.01104·SOC +0.04918 (A.25)

Rs_LPc =(0.01036 ·e0.295·SOC +3.829e −6 ·e8.363·SOC )× (

0.2304 ·Cr2 −0.775 ·Cr +1.1458

)× (−0.2068 ·SOC 2 +0.1532 ·SOC +1.011)

(A.26)

Cs_LPc = 1451 ·e−0.3283·SOC −9.562e −8 ·e22.43·SOC (A.27)

A.4. RESULTS AND DISCUSSION 189

0 0.2 0.4 0.6 0.8 1SOC

0

0.1

0.2

0.3

Ro v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

0.05

0.1

0.15

0.2

Rs v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

50

100

150

200

Cs v

s fit

[Far

ad]

0 0.2 0.4 0.6 0.8 1SOC

0

0.2

0.4

0.6

Rl v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

1000

2000

3000

4000

Cl v

s fit

[Far

ad]

C 0.05 0.05 fit C 0.1 0.1 fit C 0.15 0.15 fit C 0.25 0.25 fit C 0.3 0.3 fit C 0.4 0.4 fit C 0.5 0.5 fit

a)

c)

e)

b)

d)

Figure A.9: The values of RC elements and their fitted results: VRLA batteries during discharge.

Rl_LPc =[0.9515+0.05887 · cos (xw)−0.06694 · si n (xw)−0.01243 · cos (2xw)

−0.04918 · si n (2xw)−0.02577 · cos (3xw)−0.01927 · si n (3xw)]

× (7.677 ·e−28.9·Cr +1.7054 ·e−2.789·Cr

)× (−0.1976 ·SOC +1.169)

x = SOC ; w = 4.494 (A.28)

Cl_LPc =[1.672e4+7346 · cos (xw)+6107 · si n (xw)+3910 · cos (2xw)

+859.6 · si n (2xw)+1356 · cos (3xw)−4264 · si n (3xw)]

× (−1.272 ·Cr2 +2.346 ·Cr +0.5817

)× (−0.08145 ·SOC 2 +0.06667 ·SOC +0.9769)

x = SOC ; w = 4.76 (A.29)

DISCHARGE

Ro_LPd = 0.04153·SOC 4−0.09593·SOC 3+0.07794·SOC 2−0.0273·SOC +0.05125 (A.30)

Rs_LPd =(−0.0325 ·SOC 3 +0.06854 ·SOC 2 −0.04985 ·SOC +0.02432)

× (1.084 ·e−0.534·Cr +0.4643 ·e−14.45·Cr

)× (0.03161 ·SOC 2 −0.07195 ·SOC +1.032

)(A.31)

190 APPENDIX A

0 0.2 0.4 0.6 0.8 1SOC

0.03

0.04

0.05

0.06

0.07R

o v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

0.01

0.02

0.03

0.04

Rs v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

500

1000

1500

2000

Cs v

s fit

[Far

ad]

0 0.2 0.4 0.6 0.8 1SOC

0

0.2

0.4

0.6

Rl v

s fit

[Ohm

]0 0.2 0.4 0.6 0.8 1

SOC

0

2

4

6

Cl v

s fit

[Far

ad]

104

C 0.05 0.05 fit C 0.1 0.1 fit C 0.15 0.15 fit C 0.25 0.25 fit C 0.3 0.3 fit C 0.4 0.4 fit C 0.5 0.5 fit

a) b)

c) d)

e)

Figure A.10: The values of RC elements and their fitted results: LFP batteries during charge.

Cs_LPd = 909 ·e0.4785·SOC −764.6 ·e−7.692·SOC (A.32)

Rl_LPd =(219.3 ·SOC 8 −1031 ·SOC 7 +1986 ·SOC 6 −2021 ·SOC 5

+1167 ·SOC 4 −381 ·SOC 3 +65.8 ·SOC 2 −5.111 ·SOC +0.1754)

× (1.307 ·e−1.872·Cr +6.3160 ·e−20.67·Cr

)× (0.04198 ·SOC +0.9656) (A.33)

Cl_LPd =(1.22e4−2988 · cos (xw)−2455 · si n (xw)−1326 · cos (2xw)

−3567 · si n (2xw)−5139 · cos (3xw)−1171 · si n (3xw))

× (−2.322 ·Cr2 +2.473 ·Cr +0.5983

)x = SOC ; w = 5.522 (A.34)

COULOMBIC EFFICIENCY

The measured and fitted Coulombic efficiency result of VRLA battery is plotted in FigureA.12.

The Coulombic efficiency is expected to decrease when current increases. However,the efficiency measured with 0.2C is higher than that measured with 0.1 C, and theresult measured with 0.4 C is larger than the one measured with 0.3 C. This is because0.1 C and 0.3 C measurements were operated on the same battery, while a differentbattery was tested with 0.2 C and 0.4 C. The differences between batteries may lead to

A.4. RESULTS AND DISCUSSION 191

0 0.2 0.4 0.6 0.8 1SOC0.02

0.04

0.06

0.08

Ro v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC0

0.01

0.02

0.03

0.04

Rs v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC0

500

1000

1500

Cs v

s fit

[Far

ad]

0 0.2 0.4 0.6 0.8 1SOC

0

0.2

0.4

0.6

Rl v

s fit

[Ohm

]

0 0.2 0.4 0.6 0.8 1SOC

0

1

2

3

4

Cl v

s fit

[Far

ad]

104

C 0.05 0.05 fit C 0.1 0.1 fit C 0.15 0.15 fit C 0.25 0.25 fit C 0.3 0.3 fit C 0.4 0.4 fit C 0.5 0.5 fit

a) b)

c) d)

e)

Figure A.11: The values of RC elements and their fitted results: LFP batteries during discharge.

this consequence. Nevertheless, the Coulombic efficiency tested with higher C-rate islower than one tested with lower C-rate for the same battery. Therefore, the Coulombicefficiency is fitted into the modified Equation A.10, which was introduced in Section A.3.6.The result is shown in Equation A.35.

ηc = 0.977

[1−e

5.466

5.569e−3 Cr0.2 +0.03745

×(SOC−1)]

(A.35)

where Cr is the C-rate.

A.4.2. SIMULATION AND VALIDATIONIn this section, the simulation results and the experimental data are compared. Thecomparison of measured voltage and simulated voltage for both battery technologies areplotted in Figures A.13 and A.14.

As stated in Section A.2.2, there are three essential features that should be reflectedin the model: the storage feature, the voltage response and the parasitic side reaction.In this study, the SOC is the independent variable, and the voltage response is the mainfeature being modelled. The Coulombic efficiency is included in the storage circuit aswell as the voltage response circuit. Hence, the voltage is selected to evaluate the accuracyof the model. To quantify the model accuracy, all the modelled and measured voltagevalues are calculated and evaluated using Equation A.36. Then, for each battery, there isan averaged error value achieved among the whole charge/discharge process.

error = Vsim −Vexp

Vexp×100 (A.36)

192 APPENDIX A

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1SOC

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1C

oulo

mbi

c ef

ficie

ncy

0.1C

0.1C fit

0.2C

0.2C fit

0.3C

0.3C fit

0.4C

0.4C fit

0.5 0.6 0.7 0.8 0.9 1

0.9

0.92

0.94

0.96

0.98

0.94 0.95 0.96 0.97 0.98

0.88

0.9

0.92

0.94

0.96

0.98

Figure A.12: Measured Coulombic efficiency of VRLA battery with SOC and with different C-rate.

where Vsim is the simulated voltage and Vexp is the measured voltage.The constant voltage (CV) charging stages were omitted from the calculation because,

in the real measurement, the pause steps are accomplished by the tester, and it dependson the natural battery features. This step counting outcome accomplished by the testerin CV stage is unable to be repeated with the simulation. How it works is reflected inFigures A.13 and A.14. For example, the comparison results of VRLA batteries is shown inFigure A.13. The discrepancy in this figure between the experimental and the simulationresults (e.g., 600–1000, 250–700 and 150–650 min in Figure A.13a–c, respectively) aredue to the step counting method difference. Therefore, the CV stages for both batterytechnologies are eliminated.

The comparison results of VRLA batteries are in Figure A.13. The simulated voltagecurve increases slower than the measured one during the last charging stage before thebattery reaches the CV stage, which is, e.g., around 600 min in Figure A.13a. This slowerincrease trend of voltage is due to the modelled Coulombic efficiency loss. This can beattributed to the fact that the number of measured Coulombic efficiency data points whenthe SOC > 98% are insufficient. The fitted Coulombic efficiency result then cannot modelthe VRLA battery behaviour accurately when it reaches the high SOC stage, especiallywhen the SOC higher than 98%.

For VRLA batteries, the maximum averaged error among all eight different C-ratesis 1.7664%, while discounting the CV charging stage. The Coulombic efficiency fittingresults might be one of the reason causing the error here.

Figure A.14 shows the comparison results of LFP batteries. For LFP batteries, themaximum averaged error among all different C-rates, which is eight batteries in total, is1.6708%. The discrepancy of the wedges shaped charge/discharge steps is mainly due tothe differences in ohmic resistance. The error from ohmic resistance counts at least 0.5%.

The dynamic models of both battery technologies are therefore considered accurate.

A.5. CONCLUSIONS 193

0 200 400 600 800 1000 1200 1400 1600 1800 Time [min]

1.6

1.8

2

2.2

2.4

Vol

tage

[V]

ExpSim

0 100 200 300 400 500 600 700 800 900 1000 1100 Time [min]

1.6

1.8

2

2.2

2.4

Vol

tage

[V]

ExpSim

0 100 200 300 400 500 600 700 800 Time [min]

1.6

1.8

2

2.2

2.4

Vol

tage

[V]

ExpSim

b)

a)

c)

Figure A.13: Comparison of experimental result and simulation result-VRLA battery.: (a) test with 0.1 C; (b) testwith 0.25 C; and (c) test with 0.5 C.

RECOMMENDATIONS

Firstly, it is recommended to identify and develop a more accurate methodology tomeasure the Coulombic efficiency. The methodology presented in this chapter is sufficientover most of the SOC range, except beyond 98%, where the achievable SOC granularity ofthe method is lower than desirable. Even though more accurate industrial test methodsfor Coulombic efficiency measurement are mostly chemical-based, it is still worthwhiledeveloping a simple method with sufficient accuracy over the entire operational SOCrange, especially for electrical models. Secondly, the batteries’ dynamic behaviour wasmodelled only when the batteries were new. The dynamic performance evolution throughageing is currently ignored. This is proposed to be improved in the future work. Thirdly,the impact of temperature needs to studied as an independent and controllable variable.Finally, this EECM is proposed to be applied in actual SHS use-cases by using batterycharging and discharging profiles from SHS applications.

A.5. CONCLUSIONSIn this study, two battery technologies (VRLA and LFP) were chosen for experimentallydetermining the dynamic behaviour of the battery, with the eventual goal of constructingan EECM model. The constructed model works well throughout the C-rate range from0.05 C to 0.5 C, with an error lower than 2% for both LFP and VRLA batteries. The chosen

194 REFERENCES

0 200 400 600 800 1000 1200 1400 1500Time [min]

2.5

3

3.5 V

olta

ge [V

]

ExpSim

0 100 200 300 400 500 600 700Time [min]

2.5

3

3.5

Vol

tage

[V]

ExpSim

0 50 100 150 200 250 300 350 400 450Time [min]

2.5

3

3.5

Vol

tage

[V]

ExpSim

a)

b)

c)

Figure A.14: Comparison of experimental result and simulation result-LFP battery: (a) test with 0.1 C; (b) testwith 0.25 C; and (c) test with 0.5 C.

C-rate range gave a better granularity for the intended application of SHS. The non-linear relation between SOC and VOC, the use of 2nd order RC circuit in EECM, and thequantification of the parasitic branch of VRLA cells in the form of SOC and C-rate-basedCoulombic efficiency ηc were identified as methodological contributions for modellingthe dynamic behaviour of VRLA batteries. Additionally, the model developed in this workis at the cell level, which is fully scalable to battery pack level. The proposed model canbe used by SHS designers for understanding application-specific battery behaviour in thecontext of system design and analysis.

REFERENCES[1] IEA, Energy Access Outlook 2017 - From Poverty to Prosperity. Organization for

Economic Cooperation and Development, International Energy Agency, 1 ed., 2017.

[2] N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, andM. Zeman, “Estimating battery lifetimes in solar home system design using a practi-cal modelling methodology,” Applied Energy, vol. 228, pp. 1629 – 1639, 2018.

[3] J. Leadbetter and L. G. Swan, “Selection of battery technology to support grid-integrated renewable electricity,” Journal of Power Sources, vol. 216, pp. 376–386,2012.

REFERENCES 195

[4] P. Moseley and D. Rand, “Chapter 1 - the valve-regulated battery — a paradigm shiftin lead-acid technology,” in Valve-Regulated Lead-Acid Batteries (D. Rand, J. Garche,P. Moseley, and C. Parker, eds.), pp. 1 – 14, Amsterdam: Elsevier, 2004.

[5] N. S. Hudak, “4 - nanostructured electrode materials for lithium-ion batteries,” inLithium-Ion Batteries (G. Pistoia, ed.), pp. 57 – 82, Amsterdam: Elsevier, 2014.

[6] B. Nykvist and M. Nilsson, “Rapidly falling costs of battery packs for electric vehicles,”Nature Climate Change, vol. 5, no. April, pp. 329—-332, 2015.

[7] C. Zhang, K. Li, S. McLoone, and Z. Yang, “Battery modelling methods for electricvehicles - A review,” 2014 European Control Conference, ECC 2014, pp. 2673–2678,2014.

[8] P. Mohanty and M. Gujar, PV Component Selection for Off-Grid Applications, pp. 85–106. Cham: Springer International Publishing, 2016.

[9] D. G. für Sonnenenergie. London: Earthscan, english ed., 2005.

[10] A. Stroe, D. Stroe, M. Swierczynski, R. Teodorescu, and S. K. Kær, “Lithium-ionbattery dynamic model for wide range of operating conditions,” in 2017 InternationalConference on Optimization of Electrical and Electronic Equipment (OPTIM) 2017 IntlAegean Conference on Electrical Machines and Power Electronics (ACEMP), pp. 660–666, May 2017.

[11] B. Y. Liaw, G. Nagasubramanian, R. G. Jungst, and D. H. Doughty, “Modeling oflithium ion cells—a simple equivalent-circuit model approach,” Solid State Ionics,vol. 175, no. 1, pp. 835 – 839, 2004. Fourteenth International Conference on SolidState Ionics.

[12] S. Tian, M. Hong, and M. Ouyang, “An experimental study and nonlinear modelingof discharge I-V behavior of valve-regulated lead-acid batteries,” IEEE Transactionson Energy Conversion, vol. 24, no. 2, pp. 452–458, 2009.

[13] Introduction, pp. 1–9. Dordrecht: Springer Netherlands, 2008.

[14] T. Crompton, “1 - introduction to battery technology,” in Battery Reference Book(T. Crompton, ed.), pp. 1 – 64, Oxford: Newnes, third edition ed., 2000.

[15] C. A. Vincent, “2 - theoretical background,” in Modern Batteries (Second Edition)(C. Vincent and B. Scrosati, eds.), pp. 18 – 64, Oxford: Butterworth-Heinemann,second edition ed., 1997.

[16] M. Dubarry, V. Svoboda, R. Hwu, and B. Yann Liaw, “Incremental Capacity Analysisand Close-to-Equilibrium OCV Measurements to Quantify Capacity Fade in Com-mercial Rechargeable Lithium Batteries,” Electrochemical and Solid-State Letters,vol. 9, no. 10, pp. A454–A457, 2006.

[17] M. Root, The TAB Battery Book An In-Depth Guide to Construction, Design, and Use.McGraw-Hill Companies, 2010.

196 REFERENCES

[18] D. Doerffel, Testing and Characterisation of Large High-Energy Lithium-Ion Batteriesfor Electric and Hybrid Electric Vehicles. PhD thesis, 2007.

[19] W. Waag, S. Käbitz, and D. U. Sauer, “Experimental investigation of the lithium-ion battery impedance characteristic at various conditions and aging states and itsinfluence on the application,” Applied Energy, vol. 102, pp. 885–897, 2013.

[20] L. Lam, P. Bauer, and E. Kelder, “A practical circuit-based model for Li-ion batterycells in electric vehicle applications,” in Telecommunications Energy Conference(INTELEC), 2011 IEEE 33rd International, pp. 1–9, IEEE, 2011.

[21] C.-J. Zhan, X. Wu, S. Kromlidis, V. Ramachandaramurthy, M. Barnes, N. Jenkins, andA. Ruddell, “Two electrical models of the lead–acid battery used in a dynamic voltagerestorer,” IEE Proceedings - Generation, Transmission and Distribution, vol. 150, no. 2,p. 175, 2003.

[22] M. Ceraolo, “New Dynamical Models of Lead – Acid Batteries,” IEEE Transactions onPower Systems, vol. 15, no. 4, pp. 1184–1190, 2000.

[23] X. Hu, S. Li, and H. Peng, “A comparative study of equivalent circuit models for Li-ionbatteries,” Journal of Power Sources, vol. 198, pp. 359–367, 2012.

[24] S. Panchal, J. Mcgrory, J. Kong, R. Fraser, M. Fowler, I. Dincer, and M. Agelin-Chaab,“Cycling degradation testing and analysis of a lifepo4 battery at actual conditions,”International Journal of Energy Research, vol. 41, no. 15, pp. 2565–2575, 2017.

[25] W. E. M. Jones and D. O. Feder, “Behavior of vrla cells on long term float. ii. the effectsof temperature, voltage and catalysis on gas evolution and consequent water loss,”in Proceedings of Intelec’96 - International Telecommunications Energy Conference,pp. 358–366, Oct 1996.

[26] S. Buller, Impedance-based simulation models for energy storage devices in advancedautomotive power systems. Phd thesis, RWTH Aachen University, 2003.

[27] R. Kaushik and I. G. Mawston, “Coulombic efficiency of lead/acid batteries, particu-larly in remote-area power-supply (RAPS) systems,” Journal of Power Sources, vol. 35,no. 4, pp. 377–383, 1991.

[28] D. Andre, M. Meiler, K. Steiner, C. Wimmer, T. Soczka-Guth, and D. U. Sauer, “Charac-terization of high-power lithium-ion batteries by electrochemical impedance spec-troscopy. I. Experimental investigation,” Journal of Power Sources, vol. 196, no. 12,pp. 5334–5341, 2011.

[29] MACCOR, “Series 4000 Automated Test System,” pp. 1–4.

[30] Enersys, “Application Manual of CYCLON AGM single cell,” no. December, 2008.

[31] A123 Systems, “Nanophosphate High Power Lithium Ion Cell ANR26650M1-B,” pp. 2–3, 2012.

REFERENCES 197

[32] J. H. Yan, H. Y. Chen, W. S. Li, C. I. Wang, and Q. Y. Zhan, “A study on quick chargingmethod for small VRLA batteries,” Journal of Power Sources, vol. 158, no. 2 SPEC. ISS.,pp. 1047–1053, 2006.

[33] J. H. Aylor, A. Thieme, and B. W. Johnson, “A Battery State-of-Charge Indicator forElectric Wheelchairs,” IEEE Transactions on Industrial Electronics, vol. 39, no. 5,pp. 398–409, 1992.

[34] S. P. S. Pang, J. Farrell, J. D. J. Du, and M. Barth, “Battery state-of-charge estimation,”Proceedings of the 2001 American Control Conference. (Cat. No.01CH37148), vol. 2,pp. 1644–1649, 2001.

[35] Y. Hu, S. Yurkovich, Y. Guezennec, and B. J. Yurkovich, “Electro-thermal batterymodel identification for automotive applications,” Journal of Power Sources, vol. 196,no. 1, pp. 449–457, 2011.

APPENDIX B: EVALUATING THE

IMPACT OF TEMPERATURE ON

SOLAR HOME SYSTEMS

B.1. INTRODUCTIONAs price is a key sensitive parameter especially for rural off-grid markets, any degradationin performance or lifetime of SHS will adversely impact the overall system costs. In mostof the rural off-grid areas that are un(der)-electrified, the ambient temperatures are oftenmuch higher than the standard testing conditions (STC) temperature of 25C. Therefore,the work described in this appendix chapter investigates the impact of temperature onthe performance and lifetime of SHS. This study answers the following questions in thecontext of rural electrification:

1. What is the performance-degradation in SHS energy yield due to temperature?

2. What is the impact of temperature on the lifetime-degradation of (a lead-acid)battery in SHS?

B.2. BACKGROUNDThe block diagram of a typical SHS has been seen in Chapter 1 in Figure 1.7. We focus onPV and battery within the scope of this study. While the temperature impact on PV andbattery are quantitatively analyzed, the temperature impact on the power electronics isalso qualitatively commented upon in brief.

B.2.1. PHYSICAL EFFECTS OF TEMPERATURE ON SHS COMPONENTSTemperature is a key parameter that influences both performance (efficiency) as well asthe lifetime of SHS components, especially for implementations in off-grid areas expe-riencing high ambient temperatures. A brief technological background on the physicaleffects of temperature on SHS is given below.

INFLUENCE OF TEMPERATURE ON PV MODULE

At the device level, the temperature has two contradictory effects on a solar cell. Onthe one hand, an increase in temperature decreases the bandgap and consequently the

This appendix chapter is based on the following publication: N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer and M. Zeman, "A modeling methodology to evaluate the impact of temperature on Solar HomeSystems for rural electrification", 2018 IEEE International Energy Conference (ENERGYCON), Limassol, Cyprus,2018, pp. 1-6. doi: 10.1109/ENERGYCON.2018.8398756.

199

200 APPENDIX B

open circuit voltage (VOC) of the solar cell [1, 2]. On the other hand, having a positivetemperature coefficient of short circuit current (ISC), an increase in temperature causesan increase in the ISC. For the same input irradiance level, the loss in voltage is thedominating effect, eventually leading to a decrease in the output power [1, 2].

In terms of lifetime, crystalline silicon (c-Si) modules, the most commonly usedPV technology in SHS, have been reported to experience a lifetime-degradation rate ofaround 0.5-0.9%/year range, including the degradation due to higher temperatures in hotand humid climates [3]. The temperature-induced lifetime-degradation of PV modules isas such not explored in this study.

INFLUENCE OF TEMPERATURE ON BATTERY

Temperature can have a very short-term positive influence on the battery performance.This is because depending on the state of health (SOH) of the battery, the increase inelectrolyte temperature can lead to a heightened chemical activity for the forward reaction.If the efficiency is purely looked upon as the ratio between energy taken out to the energyput into the battery, higher efficiency can be temporarily attained as compared to thecase without any temperature increase [4].

However, an increase in temperature can be severely detrimental to the lifetime of thebattery. It can lead to accelerated failure mechanisms due to grid corrosion and waterloss for evaporation or hydrogen evolution at the negative plates of a lead-acid battery [5].

INFLUENCE OF TEMPERATURE ON POWER ELECTRONICS

Power electronics present in SHS in the form of converters generally comprise of switchesand passive elements. In terms of performance, the largest impact of temperature is onswitch conduction losses due to an increase in the on-resistance (RDSON ). In terms ofpassives, the filter capacitor value is hardly affected by temperature, while its EquivalentSeries Resistance (ESR) decreases with increase in temperature, leading to a reduction inits impedance[6]. With rising temperatures, the ferrite cores generally exhibit marginallydecreasing losses until around 80-90C, after which the losses increase [7].

In terms of lifetime, the degradation in life-cycles is usually denoted as a functionof the thermal stress cycles that the power electronic component undergoes. Very hightemperatures can lead to the copper and tin at the solder joints forming a brittle alloy, thusoften making the solder joints the weakest link in the reliability of the power electronicsin use. In any case, for the currently rated power levels in SHS (sub 200 W), the state-of-the-art power converters are expected to last at least 20 years.

B.2.2. SCOPE OF THIS STUDYThe temperature impact on SHS components can be broadly classified as either performance-degrading (drop in efficiency or energy yield) or lifetime-degrading (drop in life-cycles).Based on the relative impact of the type of degradation, the work described in this chap-ter focuses on quantifying the performance-degradation of PV modules and lifetime-degradation of the battery in SHS.

The physical effects of temperature outlined in Section B.2.1 vary in degree betweenthe kind of technology being used. Based on the application space of rural electrification,the chosen technologies for this study are crystalline silicon for the PV module, and valve-

B.3. METHODOLOGY 201

regulated lead-acid (VRLA) lead-acid gel for the battery, as these were the most populartechnologies being used by SHS providers at the time of conducting this study.

Additionally, while past studies have individually looked into the temperature impacton PV [3, 1], battery [5], there is a lack of a comprehensive work on temperature impact onSHS in the context of rural electrification and particular application use-case for a givenload profile and battery usage. This work described here aims to create fresh insights inthis direction.

B.3. METHODOLOGYThe methodology consists of simulating models for quantifying the performance andlifetime-degradation of SHS components. Inputs to the simulation and the models usedare described below.

B.3.1. INPUTS TO THE SHS SIMULATIONThe system level simulation was run over a period of 1 year corresponding to the avail-ability of meteorological data with 1-minute resolution.

METEOROLOGICAL DATA

The meteorological data was obtained from the Meteonorm software and consisted ofPV irradiance, wind speed, and ambient temperature with 1 min resolution [8]. The sunposition data was obtained using the sun position algorithm from Sandia National Labs[9]. As ground level meteorological data for an off-grid rural location could not be found,the choice of location was considered to be Ahmedabad in India (23.03N, 72.58E), acity that also receives a very high amount of irradiance (around 5.5 equivalent sun hoursper day) and experiences very high temperatures of around 40C. This is considered as agood representative choice for investigating the temperature impact on SHS.

LOAD PROFILE

Given the recent proliferation of efficient off-grid DC appliances, only DC loads wereconsidered to be used in the load profile [10, 11, 12]. The load profile was selected froma study by the authors [13], which detailed the construction of stochastic load profilesto map the energy consumption for the various tiers of the multi-tier framework formeasuring electricity access. A tier 3 household electricity need was assumed for thisstudy. Consequently, a tier 3 load profile was used as an input to the SHS simulation.

PV MODULE

The selected PV module is Jinko Solar JKM265P. Given the rating of 265 Wp, 1 PV module issufficient to cover the average daily energy needs. To satisfy the load completely, however,an appropriately sized battery storage will be needed, which is discussed in Section B.3.3.The main module characteristics have been tabulated in Table B.1.

BATTERY

A VRLA battery was chosen for the simulation. The datasheet provided in [14] was vital toperform the analysis of temperature impact on lifetime-degradation, as described later inSection B.3.4.

202 APPENDIX B

Table B.1: Main parameters of PV module Jinko Solar JKM265P.

Parameters Value

Power Pmpp (Wp) 265Area AM (m2) 1.6368VOC (V) 38.6ISC (A) 9.03Vmpp (V) 31.4Impp (A) 8.44k (%/°C) -0.41NOCT (C) 45ηstc (%) 16.19

B.3.2. ESTIMATING PV YIELDEstimating the PV yield while accounting for the thermal losses comprises multiple steps,as described below.

PLANE OF ARRAY (POA) IRRADIANCE

First, the optimal module orientation is evaluated by examining the effect of variousmodule orientations, i.e., tilts and azimuths, by calculating the POA irradiance at eachorientation. The orientation maximizing the annual POA irradiation is chosen as theoptimal one. The 3 different components of irradiance, Direct Normal Irradiance (DNI),Diffused Horizontal Irradiance (DHI), and Global Horizontal Irradiance (GHI), whichare obtained from the Meteonorm database, contribute differently at different moduleorientations. These are taken into account using the following equations, which helpcalculate the irradiance components, viz., direct, diffused, and albedo radiation incidenton the POA [2]:

cosθ = sin(Es)cosφ+cos(Es)sinφcos(Am − As) (B.1)

Gdirect = DNIcosθ (B.2)

Gdiffused = DHI

(1+cosφ

2

)(B.3)

Galbedo = GHI

(1−cosφ

2

)×α (B.4)

Where: Es: Elevation of the sun, As: Azimuth of the sun, Am: Azimuth of the module, φ:Module tilt, θ: Angle between DNI and the normal to the POA, α: albedo of the landscape(assumed 0.15). Finally, the POA irradiance is calculated as shown in Equation B.5.

GPOA =Gdirect +Gdiffused +Galbedo (B.5)

PV MODULE TEMPERATURE

As the meteorological data only gives ambient temperature, the module temperatureneeds to be then correctly estimated. Three different PV module temperature estimation

B.3. METHODOLOGY 203

models from the literature were used in this study. These are the nominal operatingcell temperature (NOCT) model, Duffie-Beckman (DB) model, and the fluid-dynamics(FD) model. Note that the cell temperature is assumed to be the same as the PV moduletemperature when referring to the quantity TM in these models.

NOCT MODEL

This is a simplified steady-state model with NOCT of the PV module as a reference point.As the NOCT for a typical c-Si module is around 40-45C, the module temperature isalways greater than the ambient temperature for a positive value of POA. The governingequation is as follows [2].

TM = Tamb +TNOCT −20C

800×GPOA (B.6)

Where TM is the module temperature, Tamb is the ambient temperature, TNOCT is theNOCT of the PV module, which is test-measured at 800 W/m2 and 20C. However, theNOCT model fails to take into account several other parameters.

DUFFIE-BECKMAN

The Duffie-Beckman (DB) model takes into account additional parameters, viz., windspeed, cell efficiency, transmittance, and absorptivity [15].

TM = Tamb + TNOCT−20C800 ×GPOA

( 9.55.7+3.8×w

)(1− ηcell

T×a

)(B.7)

Where w is the windspeed, ηcell is the cell efficiency, T is the transmittance of the modulefront, and a is the absorptivity of the module.

FLUID-DYNAMIC MODEL

Both NOCT and DB models fail to take into account the location dependent diversemeteorological conditions [2]. The fluid-dynamic (FD) model, based on a detailed thermalenergy balance between a tilted module and its surroundings, overcomes the drawbacksof both DB and NOCT models. The governing equation is stated in Equation B.8 [2]. Adetailed derivation is present in [16].

TM = aGPOA +hcTa +hr,skyTsky +hr,grTgr

hc +hr,sky +hr,gr(B.8)

Where aGPOA is the heat received from the sun; hc is the convective heat transfer co-efficient between the module and the surroundings; hr,sky and hr,gr are the linearizedcoefficients of radiative heat exchange between the upper part of the module and thesky, and the back of the module and the ground, respectively [2]; Ta, Tsky, and Tgr aretemperatures of the surroundings, sky, and ground respectively.

DYNAMIC EFFICIENCY AND PV POWER

As the output of the PV module varies based on meteorological conditions like irradiance,temperature, and wind speed, the instantaneous efficiency is dynamic and can be verydifferent from the STC efficiency. To evaluate the dynamic efficiency and PV power,

204 APPENDIX B

first the PV output power at 25C is calculated as a function of irradiance as shown inEquations B.9 through B.12 [2].

VOC(25C,GPOA) =VOC(STC)+ nkB T

qln

(GPOA

GSTC

)(B.9)

ISC(25C,GPOA) = ISC(STC)GPOA

GSTC(B.10)

Pmpp(25C,GPOA) = FF×VOCISC(25C,GPOA) (B.11)

η(25C,GPOA) = Pmpp(25C,GPOA)

GPOA × AM(B.12)

Where kB and q are the universal Boltzmann constant and charge constant respectively, nis the ideality factor (usually≈ 1.5 for c-Si [2]), and FF is the fill factor. Now the temperatureeffect can be added to the Pmpp by using the temperature coefficient of power (the k value)the manufacturer states in the PV datasheet (Table B.1), giving the dynamic PV outputpower as shown in Equation B.13.

Pmppdyn= Pmpp(25C ,GPOA)+kPmpp(STC)(TM −25C) (B.13)

Finally, the dynamic efficiency can be calculated as

ηdyn =Pmppdyn

GPOA × AM(B.14)

EVALUATING PV PERFORMANCE

With the dynamic PV output and efficiency known, the PV performance can be quantifiedbased on 3 metrics, viz. DC yield, corrected efficiency, and the Module Ideality Factor(MIF). The DC yield is simply the integrated PV output power over the period of 1 year(Equation B.15). The corrected efficiency is defined in Equation B.16.

E YDC =

∫ 1year

t=0Pmppdyn

d t (B.15)

ηcorr =E Y

DC∫ 1yeart=0 GPOA × AM d t

(B.16)

MODULE IDEALITY FACTOR (MIF)Module Ideality Factor is defined as an indicator of the proportion of expected PVyield actually available after accounting for the thermally induced losses for a givenPV module[17]. MIF is a module dependent parameter that is different for every moduledepending on the impact of temperature on the performance of the module. For an idealPV module with no thermally induced losses, the MIF=1.

MIF = E YDC∫ 1year

t=0 GPOA × AM ×ηstc d t(B.17)

B.3. METHODOLOGY 205

B.3.3. BATTERY SIZINGBased on the corrected PV output yield after accounting for thermal losses, and the loadprofile described in Section B.3.1, an appropriate battery size is chosen such that the loadprofile is satisfied at least a certain percent of the times in terms of the LLP, as seen inSection B.4.2.

B.3.4. ASSESSING TEMPERATURE IMPACT ON BATTERY LIFETIMEThe battery lifetime was estimated based on a non-empirical method including theinfluence of temperature in 3 main steps, as described below.

MANUFACTURER’S INFORMATION

The battery manufacturer often provides cycle-life curves as a function of 2 distinct stress-factors, viz. depth of discharge (DOD) and temperature, as seen previously in Chapter 4(Figure 4.3). Look-up functions are created based on these curves, and the cycle-lifecurves for the intermediate temperatures are interpolated using linear interpolation.Thus, any given temperature and DOD can be mapped to a particular cycle-life that thebattery would enjoy at those stress factor levels.

BATTERY USAGE FROM SHS SIMULATION

With the corrected PV output and the evaluated battery size (Section B.3.3), an SHSsimulation based on energy balance is run on MATLAB. A constant battery efficiency of90% and power converter efficiency of 95% are assumed during operation. The simulationhelps in understanding exactly the kind of cycling the battery undergoes in a year, whichis used then in conjunction with the manufacturer’s data as follows.

BATTERY LIFETIME ESTIMATION INCLUDING TEMPERATURE

Firstly, the micro-cycles of the battery are extracted from the SHS simulation based on thecurrent zero-crossings (ZCs), as explained in detail in Chapter 4. The overall average DOD(DOD) is calculated as shown in Equation B.18, where DODi and Ethri are the averageDOD and energy throughput values for every micro-cycle respectively [10].

DOD =

N∑i=1

DODi .Ethri

N∑i=1

Ethri

(B.18)

The mean temperature (T ) is similarly calculated for the active periods of the battery, asshown in Equation B.19. Tambienti is the mean ambient temperature for the particularmicro-cycle, while ti is the duration of the micro-cycles.

T =

N∑i=1

Tambienti .ti

N∑i=1

ti

(B.19)

206 APPENDIX B

Then, the cycle-life number n is obtained from the look-up functions described inSection B.3.4. Equation B.20 is then used to calculate the battery lifetime (L) in years [10],where Enom is the nominal battery energy capacity in Wh.

L = n ×DOD × 2×Enom

N∑i=1

Ethri

(B.20)

For comparison, the lifetime is estimated by keeping the ambient temperature constant at25C and considering ambient temperature fluctuations along with 2 scaled temperaturefluctuation profiles, as shown in Section B.4.2.

B.4. RESULTS AND DISCUSSION

B.4.1. PV RESULTS

POA IRRADIANCE AND PV ORIENTATION

Following the steps described in Section B.3.2, an optimal orientation of 26 tilt and 232azimuth (due North) were obtained, which maximized the POA irradiation from 5.34equivalent sun hours (ESH) per day (flat orientation) to 7.03 ESH per day for an optimalorientation. This optimal orientation was assumed for the rest of the PV analysis and theSHS simulation.

MODULE TEMPERATURE ESTIMATION

Based on the methodology for temperature estimation explained in Section B.3.2, themodule temperature was estimated using 3 different models from the literature. Theestimated module temperatures are shown in Figure B.1 for a representative day (day #95)of the year when the irradiance is the highest.

0 5 10 15 200

20

40

60

80

100

120

85C

Time [h]

Ave

rage

mo

du

lete

mp

erat

ure

[C

] AmbientNOCTDBFD

Figure B.1: Estimated module temperatures for different models used, along with the ambient temperature forday #95. The dashed line represents the maximum operating temperature of the PV module.

B.4. RESULTS AND DISCUSSION 207

The NOCT model is constantly pessimistic in estimating the module temperature,resulting in the highest temperature estimates of all the models. It can be seen that forthe chosen day with the highest irradiance, the DB model estimates a relatively highermodule temperature compared to the FD model. However, for the 1-year long simulation,the average temperature estimated by the DB model is marginally lower than that by theFD model.

Additionally, the maximum module temperature estimated by the NOCT model is86.7C, which is higher than the operational temperature limit of 85C of the PV module.Additionally, the NOCT model is very simplistically linear. Therefore, the NOCT model isdeemed inadequate to be applied under high irradiance conditions. On the other hand,the DB and FD models are found to be similarly applicable for the given meteorologicalconditions. Nonetheless, this close agreement of DB and FD model cannot be general-ized to other meteorological conditions. The battery sizing and SHS simulation wereimplemented using the PV temperatures as estimated by the FD model.

0 200 400 600

E YDC (STC)

E YDC (NOCT)

E YDC (DB)

E YDC (FD)

680

582

620

616

DC Yield (kWh/year)

Figure B.2: Annual PV yield evaluated under different predictions of module temperatures, including STC.

PV PERFORMANCE

With the module temperature evaluated through the different temperature models, thedynamic PV output was calculated as shown in Section B.3.2. Consequently, the metricsdefined in Section B.3.2 were evaluated. The comparison of PV yields is shown in Fig-ure B.2. The yield when the module temperature predicted by the NOCT model is themost pessimistic, while the DB and FD cases show similar yields. However, it must benoted that dynamically, DB and FD models lead to different module temperatures, as wasseen in Figure B.1. The corrected efficiency and the MIF as the outcome of various module

Table B.2: Corrected efficiency and MIF values under different module temperature estimations vis-a-vis theSTC case.

Metric STC NOCT DB FD

ηcorr (%) 16.19 13.85 14.77 14.67MIF (-) 1 0.856 0.912 0.906

temperature estimation models are tabulated in Table B.2. Again, the values for the case

208 APPENDIX B

0 2 4 6 8 100

0.2

0.4

0.6

0.8

1

Battery size [kWh]

LL

P[-

]AmbientNOCTDBFD

0 2 4 6 8 100

0.2

0.4

0.6

0.8

1

Battery size [kWh]

LL

P[-

]

AmbientNOCTDBFD

Figure B.3: LLP variation based on SHS simulations for various storage sizes and PV outputs due to differenttemperature estimation models.

of DB and FD models-based temperature estimations are quite close. However, this isliable to change based on a different location with different meteorological conditions.The most remarkable inference is that thermally induced losses can account for 10%losses in the DC yield in such a tropical climate.

B.4.2. BATTERY RESULTS

BATTERY SIZING

Based on the LLP optimization methodology described in Section B.3.3, the LLP wasevaluated for various storage sizes assuming the PV outputs due to the different temper-ature estimation models, as shown in Figure B.3. Except for the case with NOCT-basedPV output estimation, the rest of the cases largely show a similar LLP profile with varyingbattery sizes (magnified section in Figure B.3), meaning that the thermally induced losses,while reducing the PV yield, did not significantly alter the battery sizing for the givenscenario. It is an economically optimal choice to select the battery around the knee pointof the curve. This corresponds to an LLP of 1.8% and a battery size of 1.44 kWh assuminga corrected PV output as estimated by the FD model.

BATTERY USAGE AND LIFETIME

Based on the selected storage size, an SHS energy balance based model was simulated.The ZCs method described in Section B.3.4 provided a DOD of 0.3253. Using Equa-tion B.20, the lifetime was calculated for a constant temperature of 25C, and 3 differentcases, viz. with fluctuating ambient temperatures and 2 different scaling factors to theambient temperatures (1.2 and 1.5). Each of them results in a different usage mean tem-perature. The lifetime results are shown in Table B.3, where SF is the scaling factor appliedto the ambient temperature profile.

Including the variability in temperature profiles leads to a 2.6 to 33% drop in the

B.5. CONCLUSION 209

Table B.3: Impact of temperature on estimated battery lifetime.

T=25C ambient SF = 1.2 SF = 1.5

T (C) 25 26.4 31.6 39.5L (years) 7.8 7.6 6.6 5.2

estimated battery lifetime. The lifetime estimation methodology explained here considersonly the active period of battery operation. Depending on the battery usage for the specificload profile and PV output, the lifetime estimation is liable to change. Nonetheless, thetemperature impact on battery lifetime is quantified for the given location and loadprofile. Most importantly, the methodology described here can be applied to any locationand load profile to size the SHS and determine the battery usage while estimating thetemperature impact on PV performance and battery lifetime.

RECOMMENDATIONSIt must be noted that the temperature scaling factor used in the study for examining theimpact of temperature on the battery is not a true reflection of the range of temperaturefluctuations experienced by the battery on a daily basis. The study can be expandedthrough the addition of detailed temperature profiles, and further validating the modelsthrough temperature controlled experiments on battery behaviours.

B.5. CONCLUSIONThe modeling methodology described in this appendix chapter was used to quantitativelyanalyze the impact of temperature on the performance of the PV module and lifetimeof a lead-acid battery in SHS for a given location and load profile. Three different math-ematical models were used to estimate the PV module temperature. The NOCT modelwas deemed inadequate to estimate module temperature with very high POA irradiancevalues, which is a common occurrence in the tropical latitudes. The PV yield can drop toalmost 10% due to thermally induced losses. Although thermally induced losses reducethe PV yield, they are found to have minimal impact on the battery sizing for the consid-ered load profile. The optimal LLP-based battery sizing methodology was used to arriveat a battery size of 1.44 kWh. A non-empirical lifetime estimation methodology was usedto evaluate the battery lifetime. Temperature variations during the active battery periodsare shown to lead to a 2.6 to a maximum of 33% drop in battery lifetime as compared tothe standard condition of 25C. In conclusion, this chapter presented a methodology toestimate the thermally induced losses in an SHS for rural electrification, which can befurther used to size the SHS accordingly.

REFERENCES[1] A. Q. Jakhrani, A. K. Othman, A. R. H. Rigitand, and S. R. Samo, “Comparison of solar

photovoltaic module temperature models,” World Applied Sciences Journal, vol. 14,no. SPL ISS 3, pp. 1–8, 2011.

210 REFERENCES

[2] K. Jäger, O. Isabella, A. H. Smets, R. Van Swaaij, and M. Zeman, Solar Energy: Thephysics and engineering of photovoltaic conversion, technologies and systems. UITCambridge, 2016.

[3] D. C. Jordan, S. R. Kurtz, K. VanSant, and J. Newmiller, “Compendium of photovoltaicdegradation rates,” Progress in Photovoltaics: Research and Applications, vol. 24,no. 7, pp. 978–989, 2016. PIP-15-244.R1.

[4] J. Yang, C. Hu, H. Wang, K. Yang, J. B. Liu, and H. Yan, “Review on the researchof failure modes and mechanism for lead-acid batteries,” International Journal ofEnergy Research, vol. 41, no. 3, pp. 336–352, 2017.

[5] P. Ruetschi, “Aging mechanisms and service life of lead-acid batteries,” Journal ofPower Sources, vol. 127, no. 1-2, pp. 33–44, 2004.

[6] “Aluminum electrolytic capacitors - general technical information,” 2016.

[7] “Material specification for ferrite core - 3c95 data sheet,” 2015.

[8] Meteotest, “Meteonorm ver 7.1,” 2014.

[9] S. PVPMC, “Photovoltaic performance modeling collaborative,” 2014.

[10] N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P. Bauer, and M. Zeman,“A simple methodology for estimating battery lifetimes in solar home system design,”in 2017 IEEE AFRICON, pp. 1195–1201, Sep 2017.

[11] H. Olk and J. Mundt, “Photovoltaics for Productive Use Applications. A catalogueof DC-Appliances,” Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ)GmbH, 2016.

[12] Global LEAP, “The State of the Off-Grid Appliance Market,” Tech. Rep., 2016.

[13] N. Narayan, Z. Qin, J. Popovic-Gerber, J.-C. Diehl, P. Bauer, and M. Zeman, “Stochasticload profile construction for the multi-tier framework for household electricityaccess using off-grid dc appliances,” Energy Efficiency, Nov 2018.

[14] “Installation , commissioning and operating instructions for valve-regulated station-ary lead-acid batteries - solar battery data sheet,” 2015.

[15] J. A. Duffie and W. A. Beckman, Solar Engineering of Thermal Processes Solar Engi-neering. Hoboken, New Jersey: John Wiley & Sons, 4th ed., 2013.

[16] M. K. Fuentes, “A Simplified Thermal Model for Flat-Plate Photovoltaic Arrays,” tech.rep., Sandia National Labs., Albuquerque, NM (USA), 1987.

[17] N. Narayan, “Solar charging station for light electric vehicles: A design and feasibilitystudy,” Master’s thesis, Delft University of Technology, 2013.

APPENDIX C: DECENTRALIZED

CONTROL-SCHEME FOR

DC-INTERCONNECTED SOLAR

HOME SYSTEMS FOR RURAL

ELECTRIFICATION

C.1. INTRODUCTIONTo enable higher power levels of electricity access in an economically viable way, energysharing between these individual SHS through interconnectivity is a logical progression,as also detailed in Chapter 6. A rural dc microgrid comprising SHS has been proposed inliterature before [1, 2]. However, existing rural dc microgrid studies mainly focus on onevoltage level, e.g. 12 V [3]. The SHS interconnectivity must be implemented at a highervoltage level in order to reduce the conduction losses and cable costs. Existing controlschemes do not take into account such multi-voltage dc microgrids.

Additionally, in such an interconnected network of SHS, the battery usage needs to beappropriately controlled. Since the microgrid formation is assumed to be bottom-up, thecontrol scheme for energy sharing in the microgrid should also be decentralized.

In this appendix chapter, the state of charge (SOC) balancing in such an intercon-nected SHS-based dc microgrid is addressed. In particular, the adaptive droop-basedSOC control is extended for multiple voltage levels in a dc microgrid without any meansof active communication. This is achieved through the creation of a voltage dead-band,SOC-based droop resistances, and the use of voltage ratios in dc-dc converters.

The existing state of charge (SOC)-based droop controllers for microgrids do not takeinto account multiple voltage levels. E.g., droop control at 48 V is considered in [4, 5],while passive droop control for a 12 V SHS-based microgrid is studied in [6]. State ofcharge (SOC)-based adaptive droop control for distributed storage units in a microgridwith one voltage bus is presented in [7], which aims to balance the battery SOC across themicrogrid. However, in an SHS-based microgrid as presented in our study, SOC balancingbetween the batteries is not the primary objective.

In this chapter, for a decentralized rural dc microgrid comprising individual solarhome systems (SHS) with multiple voltage levels, viz., 48 V and 350 V, a decentralized

This appendix chapter is based on the following publication: N. Narayan, L. Mackay, B. O-Malik, Z. Qin, J.Popovic-Gerber, P. Bauer and M. Zeman, "Decentralized Control-Scheme for DC-Interconnected Solar HomeSystems for Rural Electrification", 2019 International Conference on DC Microgrids (ICDCM), Matsue, Japan,2019.

211

212 APPENDIX C

SHS microgrid

PV

BatteryLoads

small

Loads

350 V

bus

DC48 V 350 V

Figure C.1: SHS-based dc microgrid architecture showing the small intra-SHS loads being operated at 48 V,and high power loads being operated at 350 V. Each SHS component is connected to the 48 V bus via a dc–dcconverter.

control scheme to enable power sharing between the SHS is proposed. This controlscheme should not only facilitate power exchange but also allow for SOC balancing whilepreventing unnecessary (dis)charging of batteries. The proposed approach presents anovel way to share the power between SHS without the need for communication betweenthem. The SOC-based adaptive droop control enables different rates of charging anddischarging the batteries based on the dynamic SOC levels. The proposed microgridconverter control scheme passively conveys voltage-based information between thedifferent voltage levels both inside the SHS and outside, within the microgrid. Theproposed control scheme also prevents charging of batteries with energy from otherbatteries in the microgrid.

C.1.1. CHAPTER LAYOUT

This appendix chapter is organized into five sections. Section C.1 introduces this work,while Section C.2 presents the microgrid case-study. Section C.3 details the adaptive SOC-based droop control, and Section C.4 discusses the simulation results of a heterogeneousmicrogrid with 5 households using the proposed decentralized control scheme. Finally,Section C.5 concludes this work while listing the future work and recommendations.

C.2. MICROGRID CASE-STUDY

C.2.1. SHS-BASED MICROGRID

Figure C.1 presents the dual-voltage SHS-based microgrid. Small intra-household loadsare connected to the 48 V bus via the load converter. Higher power loads can be connectedto the 350 V bus. All the SHS components, viz., PV, battery, and load, have a dedicatedconverter that helps in implementing the decentralized droop control. Apart from power

C.3. CONTROL OF INTERCONNECTED SHS BATTERIES 213

conversion, the 48 V–350 V dc–dc converter also performs the crucial task of enablingcontrolled power sharing between the SHS in the meshed microgrid, as discussed inSection C.3.5.

C.2.2. SCENARIO FOR MICROGRID SIMULATIONA total of 5 interconnected households with SHS were considered, as shown in Figure C.2.It should be noted that only SHS 1 was considered to be having a high power load on the350 V network.

SHS 1

SHS 2

SHS 3

SHS 4

SHS 5

350Vdc

Figure C.2: Microgrid with 5 SHS interconnected on the 350 V line. Note that only SHS 1 has loads connecteddirectly on the 350 V line. All the SHS have smaller loads on the 48 V line within the household.

C.2.3. INPUT DATA TO THE MICROGRID MODELLoad data has a resolution of 1-minute and is based on a previous work by the authors [8].The PV generation data is modeled based on the irradiance data from the Meteonormtool for a tropical location that experiences on an average 5 equivalent sun hours per day[9]. The meteorological data resolution is also 1-minute. Each SHS is rated at 400 Wpwith a battery storage of 72 Ah at 48 V. The intra-SHS load profiles (loads on the 48 V bus)considered for each house were with slight variations, i.e., SHS 1 to SHS 5 were consideredwith a mix of high, low, and medium profiles, as defined in [8].

C.3. CONTROL OF INTERCONNECTED SHS BATTERIES

C.3.1. THE NEED FOR DEDICATED BATTERY CONTROLWhen interconnecting solar home systems, multiple batteries will also be connected.If batteries with identical battery chemistry and similar state of health are used, it maybe possible to interconnect them directly, i.e., without the dedicated battery convertershown in Figure C.1. However, this comes with several challenges. Care has to be takento equalize the voltages otherwise very high balancing currents would flow, causingunnecessary battery cycling. Additionally, very high short-circuit currents can occur

214 APPENDIX C

with many batteries connected in parallel. In this case, it is not possible to control thecharge of individual batteries. Moreover, it can be challenging to add more storage to thesystem as batteries may not match or have a different degradation and therefore differentcharacteristics.

Connecting batteries with dedicated converters can remove these challenges. Eachbattery can then be controlled independently. Different battery chemistries could becombined, and systems can be expanded. This additional flexibility has to be balancedagainst some drawbacks such as additional converter cost and power conversion lossesof these converters.

C.3.2. BATTERY CONVERTER

0 0.2 0.4 0.6 0.8 1

Rd

min

Rd

max

SOC [-]

Rd

[Ω]

chargingdischarging

Figure C.3: Concept illustration depicting SOC-dependent droop resistance Rd.

Assuming that each battery has its own converter, the question arises, how they shouldbe controlled. In general, it is desired to balance the state of charge (SOC) of the differentbatteries. However, specific objectives might justify a deviation from this usually acceptednorm. E.g., in case of backup power for important application, a battery may be desiredto be maintained at as high an SOC as possible.

The general approach in this chapter is to use only local voltage information forthe control. The basic principle will be to use some sort of adaptive droop control toimplement the various conditions that may appear. In general, the battery convertercould be controlled with droop control on the microgrid side (48 V bus), providing morecurrent to the microgrid the lower the bus voltage goes and charging with higher powerthe higher the bus voltage is.

Several considerations should be taken into account. In general, one undesirableoperation would be the charging of one battery by discharging another battery in themicrogrid. This creates additional losses, especially if power could also otherwise betaken directly from the charged battery at a later stage. Ideally, this would be taken care

C.3. CONTROL OF INTERCONNECTED SHS BATTERIES 215

of if all the droop curves cross the voltage axis at the same voltage. However, in practice,measurement precision can lead to undesired behavior. Therefore, a dead-band is addedto make sure that all batteries stop charging before the first battery starts discharging andvice versa.

In order to have a state of charge (SOC) of the batteries-based droop control in themicrogrid, the voltage droop slope is being adjusted based on the SOC. The chargingand discharging battery currents can therefore be in proportion to the battery SOC. Thisconcept is illustrated in Figure C.3. If a battery with a lower SOC is charging, the droopslope has to be less steep in order to increase the charging current compared to a batterywith a higher SOC. For discharging, it is the other way around: the battery with the lowerstate of charge has to have a steeper droop slope in order to discharge slower than afuller battery. Additionally, when a battery is full or empty the power has to be curtailed.Figure C.4 shows the various voltage/current droop profiles depending on the SOC ofbatteries.

Imin 0 Imax

Vm

inV

no

mV

max

DischargingCharging

Deadband

Ibatt [A]

Vb

att

[V]

SOC=0.8SOC=0.5SOC=0.2

Figure C.4: V-I diagram of the battery converter depicting the different droop slopes based on the SOC and theintroduced deadband.

Voltage drop in lines will result in some disturbance on the control. This results inbatteries closer to a load to see a lower voltage and contribute more power than otherbatteries that are further away. On the other hand, batteries that are closer to a sourcewill charge faster because they perceive a higher voltage. Qualitatively, however, this has apositive impact on reducing the line losses in the microgrid. If the dead-band is chosentoo small, it could happen that some batteries close to a source are charging while otherscloser to the load are discharging at the same time. This can be prevented by choosing thedead-band large enough, such that it covers the maximum resistive voltage drop expectedin the microgrid. However, this may not always be desirable as it will force the dischargingvoltages to be significantly lower, thus requiring more current and more line losses for thesame load power.

216 APPENDIX C

-15 -10 -5 0 5 10 15

Current (A)

38

40

42

44

46

48

50

52

54

56

58

Vol

tage

(V

)PV Converter 1

PV Converter 2

Battery Converter

Load Converter

Figure C.5: Illustration for the various converter operating modes for the PV, battery and load converter. Thelinear parts of the I–V plot refer to the droop mode, while the curved parts are the constant power mode.

C.3.3. PV CONVERTER

PV is connected to the dc microgrid by a dc/dc converter, as seen in Figure C.1. Undernormal operation, it will perform maximum power point tracking (MPPT). However, ifthe voltage in the microgrid rises too high, then the PV output power has to be curtailed.This can be achieved by adding a voltage droop on the PV output. The voltage droopshould only start at a voltage set point where all the batteries in the microgrid are alreadycharging at full power.

C.3.4. LOAD CONVERTER

The load converters usually just provide power to the loads as required while performingvoltage regulation if necessary. However, in case of insufficient power supply in themicrogrid, the load has to be shed. This can be realized through a voltage-based hysteresis,but if the load is controllable, it would be preferable to implement a droop to slowly reducethe power depending on the power availability. In a simple form, this demand responseshould happen if the voltage is so low that all the batteries in the microgrid are alreadyproviding maximal power.

The operation modes of the PV, battery, and load converters within the SHS have beencaptured in an I–V plot in Figure C.5. A voltage deadband between 47 and 49 V can beseen for the battery converter. The linear parts of the plot pertain to the droop mode,while the curved parts refer to the constant power operation. As seen in Figure C.5, theload converter does not instantaneously go from constant power mode to load sheddingbut instead undergoes a droop mode to gradually reduce the power. Two PV converterscan be seen in Figure C.5, which depict the capability of this control scheme to employdifferent PV converters rated at different powers with different setpoints for the droop

C.4. SIMULATION RESULTS 217

mode and the constant power mode. This demonstrates the utility of the proposeddecentralized control scheme in making use of differently rated resources within theSHS-based microgrid in a decentralized manner.

C.3.5. PRINCIPLE OF CONTROL FOR THE 48–350 V CONVERTER

The control scheme discussed in Section C.3 will work for the small 48 V DC nanogridwithin the SHS. To interconnect multiple SHS and form a dc microgrid, higher voltagesare necessary to reduce the line losses and consequently, the cable cost. This can beachieved by adding a dc–dc converter at each SHS, as also seen in Figure C.1, whichensures the SHS operating within the household at 48 V is connected to the 350 V network.The control of this dc–dc converter should facilitate the battery balancing without anycommunication.

The principle of operation is based on the voltage ratio k between the high voltage andlow voltage side and the nominal voltage ratio kn . The higher the voltage ratio k comparedto the nominal ratio kn , the more power is flowing from high voltage to the low voltageside, i.e., the SHS is taking power from the microgrid. This can be the case to either fulfillthe high-power loads of the SHS, or the small power loads or simply charge the batterywith excess power available in the rest of the microgrid. If the voltage ratio is equal to thenominal ratio (k = kn), there is no power exchange between the SHS and the microgrid.Finally, if the voltage ratio is lower than the nominal ratio (k < kn), power is flowing fromthe low voltage to the high voltage side. The converter current is proportional to k −kn .

C.4. SIMULATION RESULTS

The decentralized control scheme explained in Section C.3 and Section C.3.5 was im-plemented on Simulink for the microgrid case-study described in Section C.2. For the 5households, the power exchange was simulated over a single day, highlighting differentmodes of operation. The simulation results are presented in Figure C.6–Figure C.12. Thecurrent flow for each SHS component, viz., PV, battery, and load, within each householdSHS is shown in Figure C.6–Figure C.10.

Figure C.6: Current flow in the PV, battery, load and the 48 V bus for SHS 1 over a day.

218 APPENDIX C

The “Grid" current marked in these figures refers to the current flowing in the 48 Vdc bus. A positive grid current implies the local SHS at the 48 V bus side seeking powerfrom the microgrid side. A positive battery current implies discharging, while a negativebattery current implies charging of the battery. The PV and load only exhibit positiveand negative currents corresponding to the source and load nature of their behaviour,respectively. Note that the load currents in Figure C.6–Figure C.10 only refer to the lowerpower loads connected on the 48 V bus for each SHS. The high power load connected onthe 350 V line is seen in Figure C.12.

Figure C.7: Current flow in the PV, battery, load and the 48 V bus for SHS 2 over a day.

As described in Section C.2, only household SHS 1 is considered to be having highpower loads attached to the 350 V line in the case study. In the pre-dawn hours, the highpower load in SHS 1 needs to be fulfilled. As seen in Figure C.6–Figure C.12, batteries fromSHS 2 and SHS 5 contribute more power compared to the rest of the interconnected SHSbatteries owing to their higher initial SOC levels. The amount of transferred power alsodepends on the SOC-based droop control from the local battery converter, as describedin Section C.3.

Figure C.8: Current flow in the PV, battery, load and the 48 V bus for SHS 3 over a day.

C.4. SIMULATION RESULTS 219

The location of the installed high power load is not necessarily a bottleneck for theinterconnected SHS-based microgrid. As long as the microgrid voltage level is sufficientenough with respect to the individual voltage set points so that the local SHS componentskeep producing more power, the interconnection concept can be used to exchange powerwithin the microgrid.

Figure C.9: Current flow in the PV, battery, load and the 48 V bus for SHS 4 over a day.

The other benefit of excess power sharing can be immediately seen around 10:30 AMin the simulation, when batteries of SHS 2 and SHS 5 are already at 100% SOC and thelocal load demand in SHS 2 and SHS 5 is relatively low. Interestingly, the PV converterdoes not curtail the power production in SHS 2 and SHS 5 immediately. Instead, theexcess power is first used to charge the other batteries at full power (close to 8 A) so thatall the SHS batteries reach 100% SOC first. Only after that does the PV curtailment kick inat the local SHS PV converter.

Figure C.10: Current flow in the PV, battery, load and the 48 V bus for SHS 5 over a day.

Similarly, around 3:30 PM, the load demand in SHS 1 is higher than the local PVproduction, the excess power from the PV in other SHS is utilized. This is possible as thelocal loads in other SHS are lower than the local PV production. Thanks to this mode of

220 APPENDIX C

operation, the local batteries are kept untouched as long as the load demand within themicrogrid can be already utilized by the excess PV. This is where the battery deadbandcomes in handy to prevent unnecessary local battery discharge. The stored battery energycan, therefore, be used at a later stage after sunset.

Figure C.11: SOC for all the 5 interconnected SHS batteries within the microgrid over a day.

Figure C.12: Line currents in the 350 V line for each SHS over a day. The high power load connected on the 350 Vline is also shown in red.

C.5. CONCLUSIONSThis chapter proposed the concept of a decentralized control scheme that can enablepower sharing between solar home systems in a rural dc microgrid. Moreover, the ad-vantages of a communication-less, SOC-based adaptive droop control were presented.Introducing a deadband in the battery converter’s droop characteristics helped preventunnecessary cycling of the battery. Additionally, having the droop resistances adaptiveto the battery SOC levels ensures that a discharging battery experiences higher C-rateswhile at higher SOC levels and lower C-rates at lower SOC levels. Moreover, the power

REFERENCES 221

exchange modeled in the SHS-based microgrid demonstrates how the excess PV powercan be shared instead of curtailing when otherwise in standalone mode.

RECOMMENDATIONS AND FUTURE WORKThe work described in this chapter was limited to modeling and simulation of a microgridwith 5 independent but interconnected SHS. The model should be scaled up to largermicrogrid sizes while increasing the number of PV generation and load variation scenarios.Additionally, the decentralized control scheme can be implemented on hardware toexperimentally validate the SOC-based adaptive droop.

REFERENCES[1] S. Groh, D. Philipp, B. E. Lasch, and H. Kirchhoff, “Swarm electrification-suggesting

a paradigm change through building microgrids bottom-up,” in Developments inRenewable Energy Technology (ICDRET), 2014 3rd International Conference on the.IEEE, 2014, pp. 1–2.

[2] N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer,and M. Zeman, “Exploring the boundaries of solar home systems (shs) for off-gridelectrification: Optimal shs sizing for the multi-tier framework for householdelectricity access,” Applied Energy, vol. 240, pp. 907 – 917, 2019. [Online]. Available:http://www.sciencedirect.com/science/article/pii/S0306261919303502

[3] S. Groh, D. Philipp, B. E. Lasch, and H. Kirchhoff, “Swarm electrification: Investigatinga paradigm shift through the building of microgrids bottom-up,” in DecentralizedSolutions for Developing Economies. Springer, 2015, pp. 3–22.

[4] N. L. Diaz, T. Dragicevic, J. C. Vasquez, and J. M. Guerrero, “Intelligent distributedgeneration and storage units for dc microgrids - a new concept on cooperative controlwithout communications beyond droop control,” IEEE Transactions on Smart Grid,vol. 5, no. 5, pp. 2476–2485, 2014.

[5] M. Nasir, Z. Jin, H. A. Khan, N. A. Zaffar, J. C. Vasquez, and J. M. Guerrero, “A decen-tralized control architecture applied to dc nanogrid clusters for rural electrificationin developing regions,” IEEE Transactions on Power Electronics, vol. 34, no. 2, pp.1773–1785, 2019.

[6] H. Kirchhoff, M. Schmid, P. Adelmann, and K. Strunz, “Passive droop control in adecentralized 12v dc energy access microgrid with lead acid batteries,” in Micro Per-spectives for Decentralized Energy Supply: Proceedings of the International Conference(2015, Bangalore). Universitätsverlag der TU Berlin, 2015, p. 61.

[7] X. Lu, K. Sun, J. M. Guerrero, J. C. Vasquez, and L. Huang, “State-of-charge balanceusing adaptive droop control for distributed energy storage systems in dc microgridapplications,” IEEE Transactions on Industrial electronics, vol. 61, no. 6, pp. 2804–2815,2014.

222 REFERENCES

[8] T. D. Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber,P. Bauer, and M. Zeman, “Understanding the present and the future electricity needs:Consequences for design of future solar home systems for off-grid rural electrification,”in 2017 International Conference on the Domestic Use of Energy (DUE), April 2017, pp.8–15.

[9] Meteotest, “(software) meteonorm ver 7.1,” 2014.

LIST OF PUBLICATIONS

JOURNAL ARTICLES

Related to the PhD

11. N. Narayan, Z. Qin, J. Popovic-Gerber, J. C. Diehl, P. Bauer, and M. Zeman. Stochastic loadprofile construction for the multi-tier framework for household electricity access using off-grid DC appliances, Energy Efficiency, In Topical Collection of Journal Energy Efficiency onOff-grid Appliances and Smart Controls for Energy Access, Springer 2018. (Ch 3)

10. N. Narayan, T. Papakosta, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Ze-man. Estimating battery lifetimes in Solar Home System design using a practical modellingmethodology, Applied Energy, Volume 228, 2018, Pages 1629-1639, ISSN 0306-2619. (Ch 4)

9. N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M.Zeman. Exploring the boundaries of Solar Home Systems (SHS) for off-grid electrification:Optimal SHS sizing for the multi-tier framework for household electricity access. AppliedEnergy, 240, 2019, Pages 907-917. (Ch 5)

8. N. Narayan, A. Chamseddine, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M.Zeman. Quantifying the Benefits of a Solar Home System-Based DC Microgrid for RuralElectrification. Energies, 2019, 12(5), 938. (Ch 6)

7. Y. Yu, N. Narayan, V. Vega-Garita, J. Popovic-Gerber, Z. Qin, M. Wagemaker, P. Bauer, andM. Zeman. Constructing Accurate Equivalent Electrical Circuit Models of Lithium IronPhosphate and Lead–Acid Battery Cells for Solar Home System Applications. Energies, 2018,11(9), 2305. (Appendix A)

6. N. Narayan, M. Tagliapietra, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Zeman. Optimal mi-crogrid layout using Geographic Information System and graph theory concepts, submitted.(Ch 7)

5. N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer, and M. Zeman. The long road

to universal electrification: A critical look at present pathways and challenges, submitted.

(Ch 2)

Collaborations

4. V. Vega-Garita, L. Ramirez-Elizondo, N. Narayan, and P. Bauer. Integrating a photovoltaicstorage system in one device: A critical review. Progress in Photovoltaics: Research andApplications, 2019, 27(4), 346-370.

3. A. Shekhar, V. K. Kumaravel, S. Klerks, S. de Wit, P. Venugopal, N. Narayan, P. Bauer, O.Isabella, and M. Zeman, "Harvesting Roadway Solar Energy—Performance of the InstalledInfrastructure Integrated PV Bike Path," in IEEE Journal of Photovoltaics, 2018, vol. 8, no. 4,pp. 1066-1073.

223

224 LIST OF PUBLICATIONS

2. V. Vega-Garita, M.F. Sofyan, N. Narayan, L. Ramirez-Elizondo, and P. Bauer. Energy Manage-ment System for the Photovoltaic Battery Integrated Module. Energies, 2018, 11, 3371.

1. V. Vega-Garita, A. Hanif, N. Narayan, L. Ramirez-Elizondo, & P. Bauer. Selecting a suitable

battery technology for the photovoltaic battery integrated module. Journal of Power Sources,

2019, 438, 227011.

CONFERENCE PROCEEDINGS

Related to the PhD

11. N. Narayan, V. Vega-Garita, Z. Qin, J. Popovic-Gerber, P. Bauer and M. Zeman, "A modelingmethodology to evaluate the impact of temperature on Solar Home Systems for rural elec-trification," 2018 IEEE International Energy Conference (ENERGYCON), Limassol, Cyprus,2018, pp. 1-6. (Appendix B)

10. (Outstanding Paper Award) N. Narayan, T. Papakosta, V. Vega-Garita, J. Popovic-Gerber, P.Bauer and M. Zeman, "A simple methodology for estimating battery lifetimes in Solar HomeSystem design," 2017 IEEE AFRICON, Cape Town, 2017, pp. 1195-1201. (Ch 4)

9. N. Narayan, B. O-Malik, L. Mackay, Z. Qin, J. Popovic-Gerber, P. Bauer and M. Zeman, "Decen-tralized Control-Scheme for DC-Interconnected Solar Home Systems for Rural Electrification,"2019 IEEE International Conference on DC Microgrids (ICDCM), Matsue, Japan, 2019, pp.1-6. (Appendix C)

8. N. Narayan, Z. Qin, J. Popovic-Gerber, P. Bauer and M. Zeman, "Evaluating the techno-economic feasibility of battery technologies in the context of Solar Home Systems" EuropeanConference on Power Electronics and Applications (IEEE EPE ECCE Europe), Riga, 2018.

7. T. den Heeten, N. Narayan, J. C. Diehl, J. Verschelling, S. Silvester, J. Popovic-Gerber, P. Bauerand M. Zeman, "Understanding the present and the future electricity needs: Consequencesfor design of future Solar Home Systems for off-grid rural electrification," 2017 InternationalConference on the Domestic Use of Energy (DUE), Cape Town, 2017, pp. 8-15. (Partly Ch 2)

6. N. Narayan, J. Popovic, J. C. Diehl, S. Silvester, P. Bauer and M. Zeman, "Developing for

developing nations: Exploring an affordable solar home system design," 2016 IEEE Global

Humanitarian Technology Conference (GHTC), Seattle, WA, 2016, pp. 474-480.

Collaborations

5. V. Vega-Garita, D. De Lucia, N. Narayan, L. Ramirez-Elizondo and P. Bauer, "PV-batteryintegrated module as a solution for off-grid applications in the developing world," 2018 IEEEInternational Energy Conference (ENERGYCON), Limassol, Cyprus, 2018, pp. 1-6.

4. I. Sulaeman, V. Vega-Garita, G. R. C. Mouli, N. Narayan, L. Ramirez-Elizondo and P. Bauer,"Comparison of PV-battery architectures for residential applications," 2016 IEEE Interna-tional Energy Conference (ENERGYCON), Leuven, 2016, pp. 1-7.

3. V. Vega-Garita, S. Garg, N. Narayan, L. Ramirez-Elizondo, P. Bauer, "Testing a PV-batteryIntegrated Module Prototype", 2018 World Conference on Photovoltaic Energy Conversion(WCPEC-7)

LIST OF PUBLICATIONS 225

2. N. Narayan, J. Popovic, P. Bauer, A. Smets, M. Zeman, "A critical review of PV system designrules for optimizing energy yield and space utilization", 2016 European Photovoltaic SolarEnergy Conference (32nd EUPVSEC)

1. L. Locatelli, N. Narayan, A. Battaglia, A. Canino, C. Gerardi, O. Isabella, M. Zeman, "Correctingtemperature-dependent efficiency model for commercial double junction thin film siliconmodules", 2015, European Photovoltaic Solar Energy Conference (31st EUPVSEC)

ACKNOWLEDGEMENTS

People always warned me how a PhD project could finish the joy in one’s life. Instead, Iwas fortunate to experience the joy of finishing the PhD. Of course, several people havebeen a part of this journey, and I shall acknowledge them here in a rather haphazardorder. In case you don’t find your name or contribution here, please understand that abyproduct (amongst many others) of my PhD thesis is a fuzzy memory. Please know thatin my heart of hearts and a grateful yet forgetful mind, your contribution holds a specialplace and that recollecting your name the night before this book was sent to the printersproved unsuccessful. You know who you are, and a big thanks to you.

Firstly, I’d thank my promotors prof. Pavol Bauer and prof. Miro Zeman for giving methe opportunity to pursue my PhD on such a fulfilling problem statement. I thank Pavolfor giving me the benefit of the doubt and showing immense patience with my results,and also for the timely nudges that brought me the much-needed perspective within myPhD project. I thank Miro for the brief yet fruitful interactions throughout the PhD, alwaysbeing positive and supportive of my work.

Thanks to Dr. Zian Qin, my daily supervisor, for always being ready to have a discus-sion. Zian, though you moved to Delft while I was halfway through the PhD project, youimmediately came on board and understood and accepted the challenges such a uniquePhD project posed. Sorry for thoroughly exploiting your open-door policy, or assumingthat you had one in the first place while you perhaps didn’t.

Thanks to Dr. Jelena Popovic, my daily supervisor for the first year and a half. Youcontinued to provide invaluable feedback and kept me up to date with the ground realityof SHS scaling difficulties. Thank you for initiating the project and choosing me to workon it. Hope I was able to do some justice to the initial problem statement you had in mind.I’m very glad that you are continuing to battle the EA problem on academic as well asentrepreneurial fronts. Here’s hoping that we can make some dents with technology inthis shape-shifting EA monster.

Frans Pansier, it was always a pleasure to converse. I could discuss my project at anylevel of abstraction with you, and you always understood and appreciated the complexityand challenges of the project in all its glory. Thank you for sharing your treasure chests ofknowledge with me.

Thanks also to my mentor Prof. Inald Lagendijk for the limited but helpful interactionsin my first year. Thanks to my doctoral defense committee members, Prof. Peter Lund,Prof. Arno Smets, Dr. Roland Valckenborg, and Dr. J. C. Diehl for agreeing to be part of thecommittee. I appreciate the timely feedback and constructive discussions on my thesismanuscript.

I benefited from interactions with SHS companies during this project. Thanks toBernard and Thomas from SolarWorks! for always being enthusiastic about discussingvarious aspects related to this dynamic sector. Thanks to Jeroen Verschelling from Kam-Works for collaborating with us through an MSc thesis project in Cambodia. I’m grateful

227

228 ACKNOWLEDGEMENTS

to Ashley from BBOXX for an insightful discussion in the formative months of my PhD,and also to Iwona and Chris for helping me whenever I reached out to them. Also thanksto Hannes Kirchhoff and Neel Tamhane from ME-SolShare, Rob de Jeu and Marcel vanHeist from Rural Spark, and Vinay Jaju from ONergy for early discussions.

Thanks to TU Delft Global for their fellowship that supported this PhD. Nick, Jennifer,Esther, Annelies, Friso and team were helpful when both the initiative and my project wereat a fledgling stage. The Delft Global community at large has been very helpful throughoutmy project. Especially in the student projects of JC, there are plenty of valuable insightsthat have been useful more than once.

To go through the PhD and come out relatively sane, it’s important to have the rightset of people around you. Especially in the formative phase of a PhD, the influenceof peers in a research group can have the impact of a dynamite — both destructiveand constructive, but without the fortune of the explosives industry and the legacy ofan annual international award. And before this metaphor blows up in my face, I shallcontinue. Tsegay, thanks for your practical tips during coffee conversations. Laurens,hopefully, we can still collaborate in some form or the other; good luck with all thechallenges and (dc) opportunities. Soumya, wouldn’t want to make fun of you here, wouldbe quite the suicide, despite the anecdotes in my arsenal. We should play chess againsometime. Udai, it was always nice to have discussions on a range of topics. Nils, my lunchhunger has been conditioned over time to trigger with your ‘lunch’ pings. Conveniently,your pings for football also get me hungry for food. I blame it on the WhatsApp groupnamed after a meal. Pavel, starting with the APE lab supervision back in 2016, I’ve alwaysenjoyed discussing a variety of topics with you from arts to history, including (Except!)the Mongols.

Ibra, thanks for being the only other nut to splurge on good coffee while on a PhDstipend. Always wondered if your proclivity to dish out GIFs has anything to do with yoursinging aspirations (keyword: Tenor). Lucia, thanks for never making me feel like we hada wall between our offices — always heard you loud and clear. Between the notes youhit in your conversation and Ibra in his singing, you’ve got some unique voice signaturescoming out of that office. Faisal, thanks for reminding me how much I missed playingtennis. No thanks though for the furious yet graceful backhands that I could never return.Do keep that poet in you alive too! Marco, I hope by the time you’re done with your PhDyou’d be less traumatized by pineapples on pizzas. Francesca and Wenli, (wireless) powerto you guys! Mladen, never thought you’d start a trend with the man-purse, but you founda follower in Aditya. Aditya, props to you for always mixing serious scientific work withyour monkey business. Thanks to Ilija, Tianzhu, Jiayang, Piao, Thiago, Guillmero, Luis,Laura, Wiljan, Babak, Djurre, Farshid, Alessandro, Dhanashree, Mohamad, Armando,Jianning, Fabio — because interaction during coffee, cake, and even group meetingshelped.

In terms of the office room, Tim was the first glimpse into a focused PhD researcherbalancing family and his PhD. I realized a year later that it wasn’t an exception but thenorm in the ‘cursed’ family-room office of LB 3.680. Gautham, I’ve known you sincewe came here for our MSc. Thanks for all the discussions, plants, and delicious food(the edible sort of plants) we had in the office, as well as helping out with all the smarttips and advice to navigate through the everyday PhD/administrative hurdles. Victor -

ACKNOWLEDGEMENTS 229

Man, thanks for being an inspiration throughout my PhD. After seeing you juggle PhD,life, wife, and son, I really had no room for complaints. I’ll never forget your pearls ofwisdom and secret to a happy life ;) I’ll always cherish our chess discussions, and thehelpful crapohccino breaks. Yunhe, thanks for (briefly) being part of the office. When Isay the office benefited from your presence, I mean it, including the walls and the ceiling— you’ve definitely left your Mark. I am glad you found humans to practice martial artsagainst at the sports center.

Thanks to Minos, VGP, Dong, Martin, Milos, and others who graduated during myearly PhD years — you guys made me reaffirm this journey has an end. Special thanksto Joris, Bart, and Harrie for helping out in the lab. I’ll remember the brief time spentin the group by Juan Pablo, Na Li, Shubhangi, Andreas, Yufei, Jintao. Thanks also to thesecretaries Sharmila, Ellen, Ilona for helping out with any queries. Sharmila, specialthanks for restocking the cookies in your office from time to time. For the new researcherswho just joined the group, and for the many about to, I wish you the best.

Had the good fortune of working with several MSc students during the thesis. I learnt alot, both technically and culturally while working with them. Mentoring students was alsoan exercise in introspection. I accomplished more in the thesis while working with themthan I could ever have had by myself. Natalie, Yunhe and Thekla were the first students.Gaurav and Thomas joined from TPM and IDE, with fieldworks in India and Cambodiarespectively. I co-supervised Bryan, Wenrui, and Yunizar with other PhD researchers. Thelast set of students were Naveen, Ali, Michele, and Luca. You all made a difference to myPhD journey. Thank you all for wanting to pursue your MSc theses with me.

To my Delft Global fellows, thanks for supporting me. In you, I found a communityalso trying to balance the technological V social dimensions of your projects. Roos, Iwish I had your abilities to just figure things out so easily. Thanks for our support-groupcoffees and good luck with your final lap, almost there! Michel and Mitasha (M&M), youguys inspire me to no end. Michel, I am glad I got you back onto coffee; here’s to manymore. Dominik, must catch up on those Vietnam stories. Juan Carlo, great project —Keep it flowing and pump it up! Saqr, Henry, keep the energy high. Pieter, really like yourhappy-go-lucky attitude with which you seas the day, work or otherwise. Stay current,and keep it floating! Anteneh, Rachel, you guys are doing phenomenal work together.Yask, you’ve built a very fulfilling project for yourself, brick by brick. Tope, you are aninspiration; good luck with your company and your re-search. Juan, high five and athumbs up to your PhD project; one day I’d love to play rock, paper, scissors with yourprototype (and win). Camille, Petra, Merel, Christine, Mona, Hendrik — keep up the greatwork. And the new fellows: Monica, David, Daniel, welcome and good luck!

JC, thanks for always helping out with your extended network and infectious enthusi-asm. I do hope you will add Energy back onto your plate apart from Health. Abhigyan, weshould continue our discussions — technology alone is not going to help in SDG 7. Astrid,thank you for understanding my story so well and producing those beautiful illustrations.

Since July 2019, I’ve also found another warm work environment in 5.29 (extra warmin those summer days). Claire, thanks for always being so understanding and supportive;I’ll try my best to keep short updates during coffee breaks actually within 10 mins. Roel,thanks for letting me digress during all the bilas, and also telling me what bilas were inthe first place. Cheers to Danielle for always switching languages just for me in the office.

230 ACKNOWLEDGEMENTS

Sophie, you had me on day 1 when you said I could share your stash of goodies — bet youdidn’t figure back then you’d also be sharing an annoying amount of daily puns. Thanksfor keeping the office ‘not too old’. On that note, Petra, welcome to the team.

For all the LaTex troubles and MATLAB tips, I cannot thank the unsung heroes ofStack Exchange enough. My love-hate relationship with LaTex would have ended a longtime ago if it weren’t for you guys. Also, a shout-out to Coffee Star and Brandmeester inDelft (and the pico-baristo at home) for breaking me free of the Stockholm syndrome I’dotherwise developed towards the crapohccino machines.

For my home away from home, cheers to the Beachhouse, and the beachboys — theother family. Chris(-Helena-Emilian), Manu(-Katrin), Philipp(-Steffi-Johanne), Louis(-Stef-Walden) — how quickly the family has grown — thanks for starting gen-1 of thebeachhouse. Akshay, Danny, Paula(-Mike), and later Marija, Jo, I will forever cherish theendless food, chai, board games, armchair philosophizing, geeking out, homophones,horizontal skyscrapers, foiling robberies, ninja mom adventures, YT education, and ofcourse, buxfer emails . And the old cartel friends who still managed to stay in touchdespite the geographies — Sutji(-Peii-Clara-Cederic), Hugo(-Rene), Catherin(-Daniel-Vera), Yuan(-Yifan). Special shout out to the colony friends and their better halves, someof who are halfway across the world now — I am so fortunate to have those childhoodbonds still running strong. To my NITC corridor friends, after a bit of a hiatus, I was gladto be back in touch with you guys.

Also experienced the joy of celebrating my wedding during the PhD in 2016, both inIndia and in NL. The band of friends who made it all the way to India and were braverthan me in trying all the food — props to you guys. That Great Indian Wedding was anunforgettable experience. I’ve also been fortunate to have made great friends here over theyears. Ashish (le bro), Riddhi-Varun, Amreen-Harmen, Jefta-Jo-Ty-Dy, Gunjan-Shreyas,Ying Zi, Rishabh, Romy, Annalena-Suzanne, Akshara, Aakarsh, Esther-Vincent, Elvira-Berend, Sheng, JK, Luc-Marie-Nour, Mo-Johnny, Laura, Gintare, Anna, Leila, Delaram-Dena, and the list goes on. And the friends from the verre oosten — Nestor, Una, Jelenaand Mark — here’s to more cats, babies and bears, and as always, looking forward toseeing everyone next (will be on my Judgment Day I suppose). Also, special thanks forthe Dutch translation! Cheers to my Findian friends Arun & Amrita; from webcomics tochess, we have managed to stay in touch all these years, impressively without any IMs.

To team Harbour View — mumzie, poppy, and aafy, thank you for always making mefeel so loved and pampered, and part of an epic adventure whenever we are travelingtogether. Can’t wait for the next adventures and bezzique sessions. Mumzie, I’ve enjoyedall our conversations from history and religion to the latest in your scientific research.Poppy, I always admire your energy and spirit; I wish I could be even half as active. Aafydaffy, my munchkin, love you for always taking the right (my) side in all the board gamearguments. Better come back soon!

To team UK — bro, masi, siddu, loved my family time there when I could just lazearound. We badly need to increase the channel crossings. Thank you for always lookingout for me, bro. Knowing you had my back gave me the fortitude to always soldier on. Tomasi - I may not say this often enough, but thank you for being the silent prime moverof the team in the background. Sid, you are growing up too fast; come here soon so

Some stories could not make this official list as the narration of events is still disputed

ACKNOWLEDGEMENTS 231

we can keep track. Can’t wait to play chess with you. Love you guys. To pati, who hasunconditionally loved me since I was a child, and still worries that I’ll catch a cold in therain. I’ll forever cherish the memories with you and thatha. And mathe, you and appanever made me second-guess my decisions and supported me despite how hard it got foryou. I’ve never really had it tough in life, not because life has been easy, but because youguys were the real heroes always fighting the battles in the background, making life seemeasy. Love you, and thank you for always being there.

And finally, to Sanaa , words won’t do justice to your contribution in me accom-plishing this PhD. While that could have been an excellent excuse not to write anythingmore, I shall still endeavour. You’ve been with me through this every step of the way.You’ve been the rock — not the kind that traps someone between itself and a hard place —that I could always count on. For all the evenings I worked late, the movies we missed,the outings I couldn’t join, the weekends I couldn’t chill with you, thank you for alwaysunderstanding. For all the times you did put your foot down wanting me to take a break, Iam actually grateful (that the foot came down for my benefit). And you know how much Ilove challenges, that’s why you always planned our holidays right around my deadlines.Some of my fondest memories in these years are from the weekend getaways and thetravels. Oh, and thank you for also losing a few games every now and then just to makeme feel good amid all my work deadlines (*cough*), just like you forgive me every timeyou are wrong. But seriously, thank you for making this happen with me. You know thiscouldn’t have been possible for me without you. I wouldn’t want it any other way.

Because I know you will read this, I shall stick to facts and only facts

ABOUT THE AUTHOR

Nishant was born in the tiny town of Dombivli in theoutskirts of Mumbai, India in 1986. He grew up inthe unusually calm and scenic part of Mumbai calledAnushakti Nagar, finishing higher secondary schoolingin 2004 from Atomic Energy Junior College. He obtainedhis bachelor’s degree in Electronics and Communica-tion Engineering in 2008 from National Institute of Tech-nology Calicut, India. He then worked for 3 years in thesemiconductor industry at Texas Instruments India atthe Bangalore R&D centre as a Hardware Digital DesignEngineer.

Making up his mind to switch from semiconductors to renewable energies, he movedto the Netherlands in 2011 to pursue a Master’s degree in Sustainable Energy Technology.He successfully obtained his MSc in 2013, specializing in solar energy and storage, fullyrealizing the irony that solar PV energy is founded on semiconductors. From 2013 to2015, he worked on various research and education projects related to PV system design,massive open online course (MOOCs) development, PV Lab, smart grids, and wirelesspower transfer in the Photovoltaic Materials and Devices (PVMD) and DC Systems, En-ergy Conversion and Storage (DCES) groups within the Electrical Sustainable EnergyDepartment (ESE) of TU Delft, respectively.

In July 2015, he received a fellowship from TU Delft | Global Initiative to pursue aPhD in the DCES group on solar home systems for improving electricity access — a topicboth technologically and personally close to his heart. Until June 2019, he worked onhis PhD in the multi-faceted problem space of energy access, focusing on solar homesystems, battery storage, and decentralized DC microgrids. In the last few years, hehas met many inspiring people devoted to the cause of the UN SDGs, especially SDG 7(‘Ensure access to affordable, reliable, sustainable and modern energy for all’). Throughthe work and interactions during his PhD, he is surer than ever that the key to solvingurgent global issues lies in multi- and trans-disciplinary efforts. From July 2019, he hasbeen developing the Energy Access Program of TU Delft | Global Initiative. He is currentlybuilding a research roadmap to bring together researchers and experts cross-cuttingdifferent expertise as well as local partners in un(der-)electrified regions for co-solvingthe multilateral issues surrounding SDG 7.

233